Modelling of colour appearance of textured colours and ... · shopping online. While a smartphone...
Transcript of Modelling of colour appearance of textured colours and ... · shopping online. While a smartphone...
Elham Khodamoradi Page 1
Modelling of colour appearance of textured
colours and smartphones using CIECAM02
By
Elham Khodamordi
November 2016
School of Science and Technology
Middlesex University
London
A dissertation submitted to Middlesex University London
In partial fulfilment of the requirements for the degree of
Doctor of Philosophy
Elham Khodamoradi Page 2
Abstract
The international colour committee recommended a colour appearance model,
CIECAM02 in 2002, to help to predict colours under various viewing conditions
from a colour appearance point of view, which has the accuracy of an averaged
observer. In this research, an attempt is made to extend this model to predict
colours on mobile telephones, which is not covered in the model. Despite the
limited size and capacity of a mobile telephone, the urge to apply it to meet
quotidian needs has never been unencumbered due to its appealing
appearance, versatility, and readiness, such as viewing/taking pictures and
shopping online. While a smartphone can act as a mini-computer, it does not
always offer the same functionality as a desktop computer.
For example, the RGB values on a smartphone normally cannot be modified nor
can white balance be checked. As a result, performing online shopping using a
mobile telephone can be difficult, especially when buying colour sensitive items.
Therefore, this research takes an initiative to investigate the variations of colours
for a number of smartphones while making an effort to predict their colour
appearance using CIECAM02, benefiting both telephone users and makers. This
thesis studies the Apple iPhone 5, LG Nexus 4, Samsung, and Huawei models,
and compares their performance with a CRT colour monitor that has been
calibrated using the D65 standard, to be consistent with the normal way of
viewing online colours. As expected, all the telephones tested present more
colourful images than a CRT. Work was also undertaken to investigate colours
with a degree of texture. It was found that, on CRT monitors, a colour with a
texture appears to be darker but more colourful to a human observer. Linear
modifications have been proposed and implemented to the CIECAM02 model to
accommodate these textured colours.
Elham Khodamoradi Page 3
Acknowledgments
First and foremost I offer my sincere gratitude to my supervisors, Professor
Xiaohong Gao and Professor Richard Comely, who have supported me with their
guidance and encouragement throughout my thesis. They motivated me to get a
deep understanding of the subject with their endless patience and knowledge.
My deepest gratitude goes to my family for their unflagging love and support
throughout my life; this dissertation is simply impossible without them. I am
indebted to my father, for his care and love.
I cannot ask more from my mother, as she is the most perfect mother in the
world and her unconditional love has always been my main support throughout
my difficult times. I have no suitable words that can fully describe her everlasting
love for me and I feel proud to been born and raised in such a caring, loving and
magnificent family.
To my wonderful son Kevin, who has enlightened my world with his presence.
To all my friends, who have supported me all these years and encouraged me to
follow my dream unconsciously since I started university.
Finally, I would like to use this opportunity to thank Mrs Sarah Brook for initial
proof-reading and giving me feedback during my write up and to Mr Leonard
Miraziz who has always provided me with a computer and technical support
during all my PhD years at Middlesex University.
Thanks also go to those observers who have helped to perform numerous
psychophysical experiments.
Elham Khodamoradi Page 4
Table of Contents
Abstract........................................................................................................................................ 2
Acknowledgments .................................................................................................................... 3
1. Introduction .................................................................................................................... 14
2. Literature Review ......................................................................................................... 16
2.1 Overview of Colour Science .................................................................................. 16
2.1.1 Light Source ...................................................................................................... 18
2.1.2 Object ................................................................................................................ 19
2.1.3 Observer .......................................................................................................... 19
2.2 Colour Vision ...................................................................................................... 21
2.2.1 Human Eye Structure ........................................................................................ 21
2.2.2 Colour Vision Theory ........................................................................................ 22
2.2.3 Colourimetry ..................................................................................................... 25
2.2.4 Subject Estimation ............................................................................................ 26
2.3 Colour Space and Model ....................................................................................... 27
2.3.1 CIEXYZ ............................................................................................................... 28
2.3.2 CIE xy chromaticity diagram ............................................................................. 28
2.3.3 CIELUV ............................................................................................................... 30
2.3.4 CIELAB ............................................................................................................... 32
2.3.5 The Input and Output Information for CIECAM02 ........................................... 34
2.3.6 CIECAM ............................................................................................................. 41
2.3.7 Colour Appearance Datasets ............................................................................ 44
Elham Khodamoradi Page 5
2.4 Colour Appearance Phenomena ............................................................................. 45
2.4.1 Crispening Effect ............................................................................................... 47
2.4.2 Hunt Effect. (Colourfulness increases with luminance) ................................... 47
2.4.3 Stevens Effect (contrast increases with luminance) ....................................... 48
2.4.4 Bezold Brucke Effect ......................................................................................... 49
2.4.5 Colour Constancy .............................................................................................. 50
2.4.6 Metamerism ..................................................................................................... 51
2.5 Chromatic-adaptation models ............................................................................ 52
2.5.1 Colorimetry ....................................................................................................... 54
2.5.2 Colour-matching functions ............................................................................... 54
2.5.3 Chromaticity Diagrams ..................................................................................... 55
2.6 Experimental Method of Magnitude Estimation to Determine Colour
Appearance ................................................................................................................ 56
2.7 Summary ........................................................................................................................... 60
2.8 Literature Review on chromatic texture ............................................................... 62
2.8.1 Texture definition ............................................................................................. 63
2.8.2 Chromatic texture ............................................................................................. 63
2.9 Literature Review on Smartphones ......................................................................... 65
2.9.1 Introduction to Smartphone Global trends ...................................................... 65
2.9.2 The rise of mobile internet worldwide ............................................................. 66
2.9.3 Mobile and user expectations .......................................................................... 68
2.9.4 Smartphone UK trends ..................................................................................... 69
2.9.5 Smartphone changing photography ................................................................. 70
Elham Khodamoradi Page 6
2.9.6 Smartphone popularity ..................................................................................... 70
3. Methodology ...................................................................................................................... 73
3.1 Apparatus ................................................................................................................ 74
3.1.1 Viewing cabinet ................................................................................................ 74
3.1.2 Colour meter, CA-100 ....................................................................................... 74
3.1.3 Monitoring calibration (Spyder) ...................................................................... 75
3.2 Experimental setup .............................................................................................. 77
3.3 Subject training .................................................................................................... 78
4. Results on subject training and textured colours ................................................. 82
4.1 Evaluation of Colour Monitor .................................................................................. 82
4.2 Evaluation of Subjects’ Data.................................................................................... 82
4.3 Experiments of textures ........................................................................................... 85
4.4 User study ................................................................................................................ 88
4.5 The effect of texture on the appearance of colours ................................................ 96
4.6 Data prediction using CIECAM02 .......................................................................... 104
5. Results on smartphones ............................................................................................... 110
5.1 Summary of the experimental setting .................................................................... 110
5.1.1 Smarthones Colour Gummet .......................................................................... 115
5.1.2 Observer Variations ........................................................................................ 117
5.1.3 Setup on a Smartphone .................................................................................. 118
5.1.4 Apparatus........................................................................................................ 118
5.2 Comparison of Experimental Data ....................................................................... 121
Elham Khodamoradi Page 7
6. Modelling of iPhone Appearance using CIECAM02 ............................................. 127
6.1 iPhone data analysis ............................................................................................. 128
7. Results and Discussions ............................................................................................... 137
7.1 Testing CIECAM02 ............................................................................................... 137 7.2 Refinement of CIECAM02 ..................................................................................... 138
8. Modelling of iPhone appearance using CIECAM02 ............................................. 140
8.1 Comparison with other smartphones .................................................................... 141
9. Summary of Phase 2 ................................................................................................... 144
10. Conclusion and future work .................................................................................... 146
11. REFERENCES ................................................................................................................. 179
Elham Khodamoradi Page 8
List of Figures
Figure 2.1 Three essential components to generate a colour [1] _____________________________ 16
Figure 2.2 Three required to generate a colour [2]_________________________________________ 17
Figure 2.3 The Newton experiment [3] __________________________________________________ 18
Figure 2.4 Relationship between colour temperature and spectral power distribution [4] ________ 19
Figure 2.5 Colour image observation [5] _________________________________________________ 20
Figure 2.6 Human perception colour matching function [6] _________________________________ 21
Figure 2.7 Human eye receptor [7] _____________________________________________________ 22
Figure 2.8 Anatomy of the Human Eye [8] _______________________________________________ 22
Figure 2.9 A visual arrangement for producing a colour by mixing the light from three difference
coloured lamp [9]--------------------------------------------------------------------------------------------------------------- 24
Figure 2.10 Quantitative definition of a stimulus [10] _______________________________________ 26
Figure 2.11 Chromaticity Diagram [11] ___________________________________________________ 30
Figure 2.12 Chromaticity Diagram [12] ___________________________________________________ 31
Figure 2.13 Colour Appearance Model [13] _______________________________________________ 36
Figure 2.14 Colour perception [14] ______________________________________________________ 36
Figure 2.15 Colour perception [15] ______________________________________________________ 38
Figure 2.16 Colour perception [16] ______________________________________________________ 44
Figure 2.17 Colour perception [17] _______________________________________________________ 46
Figure 2.18. Mobile usage comparison between UK and middle Europe and USA companies [18]----- 46
Figure 2.19 Hunt effect [19] ____________________________________________________________ 46
Figure 2.20 Steven effect [20] __________________________________________________________ 46
Figure 2.21 Bezold Brucke effect [21]----------------------------------------------------------------------------------- 47
Figure 2.22 Colour constancy [22] _______________________________________________________ 48
Figure 2.23 Metamerism [23] ___________________________________________________________ 49
Figure 2.24 Colour matching diagram [24]----------------------------------------------------------------------------- 50
Figure 2.25 Chromaticity diagram [25] ___________________________________________________ 51
Elham Khodamoradi Page 9
Figure 2.26 Colour vision test [26] ______________________________________________________ 52
Figure 2.27 Number of mobile phone global users [27] ______________________________________ 55
Figure 2.28 Comparison between mobile devices and fixed devices [28] _______________________ 56
Figure 2.29 Device comparison [29] ______________________________________________________ 62
Figure 2.30 Mobile usage comparison between UK, middle Europe and USA companies [30]______ 66
Figure 2.31 Growth of UK mobile smartphone users [31] ____________________________________ 67
Figure 2.32 Colour viewing cabinet [32]-------------------------------------------------------------------------------- 68
Figure 2.33 Colour meter, CA-100 [33]-------------------------------------------------------------------------------- 69
Figure 2.34 Spyder sensor caliobration [34]------------------------------------------------------------------------- 70
Figure 3.1 Thirty test colours presented on a CIE xy Chromaticity diagram [35]-------------------------- 75
Figure 3.2 A viewing angle on 1931 CIE and 1964 CIE standard [36]----------------------------------------- 76
Figure 3.3 The paradigm of the experimental pattern [37]----------------------------------------------------- 76
Figure 3.4 The comparison results between female (x-axis) and male (y-axis) observers [38]--------- 77
Figure 3.5 Experimental patterns in the 6 colour and textured experiments [39]-------------------------- 78
Figure 3.6 Experimental pattern for Experiment 1 (no texture) [40]------------------------------------------- 78
Figure 4.1 Textured colour samples (experiment 2).[ [41]--------------------- ---------------------------------- 85
Figure 4.2 The tristimulus values (x, y) of thirty test colour samples for the six experiments
conducted in Phase 1. [42]--------------------------------------------------------------------------------------------------- 86
Figure 4.3 Comparison of estimation between Subject 1 (A) and means. [43]------------------------------- 87
Figure 4.4 Plot between mean lightness (x) with lightness estimation by subject 1 (y). [44]------------ 87
Figure 4.5 Comparison results between mean estimations of lightness ,colourfulness and
hue for Experiment 1 to6 [45] ---------------------------------------------------------------------------------------------- 88
Figure 4.6 Comparison results between Experiment 5 ,6 ,4 [46]---------------------------------------------- 89
Figure 4.7 Comparison results between Experiment 4 ,6 [47]-------------------------------------------------- 91
Figure 4.8 CIECAM02 lightness prediction compared with subjects’ estimations Experiments
1 to 6 [48]----------------------------------------------------------------------------------------------------------------------- 98
Figure 4.9 CIECAM02 colourfulness prediction compared with subjects’ estimations Experiments 1 to 6
[49]--------------------------------------------------------------------------------------------------------------------------------100
Elham Khodamoradi Page 10
Figure 4.10 CIECAM02 hue prediction compared with subjects’ estimations Experiments
1 to 6 [50]------------------------------------------------------------------------------------------------------------------------ 101
Figure 4.11 The paradigm of the experimental pattern [51]----------------------------------------------------- 105
Figure 4.12 The x, y data of the thirty sample colours plotted on the CIE xy Chromaticity diagram for
the four telephones [52]----------------------------------------------------------------------------------------------------- 106
Figure 4.13 The x, y data of the thirty sample colours plotted on the CIE xy Chromaticity diagram for
the three iPhones studied in this research. [53]---------------------------------------------------------------------- 107
Figure 5.1 The x, y data of the thirty sample colours plotted on the CIE xy Chromaticity diagram for the
LG-Nexuse4 studied in this research [54]------------------------------------------------------------------------------- 113
Figure 5.2 The x, y data of the thirty sample colours plotted on the CIE xy Chromaticity diagram for the
HuaWei studied in this research. [56]----------------------------------------------------------------------------------- 114
Figure 5.3 The x, y data of the thirty sample colours plotted on the CIE xy Chromaticity diagram for the
Samsung studied in this research. [57]--------------------------------------------------------------------------------- 115
Figure 5.4 Sample test on mobile phone screen [58]--------------------------------------------------------------115
Figure 5.5 Mobile phone under viewing cabinet position [59]------------------------------------------------ 116
Figure 5.6 Subject’s viewing position [60]--------------------------------------------------------------------------- 116
Figure 5.7 Colour meter, CA-100 [61]--------------------------------------------------------------------------------- 118
Figure 5.8 Comparison results between 2 iPhones using CIELUV L* and C* [62]---------------------------- 119
Figure 5.9 Comparison of CIELUV L* and C* between the SamSung phone and CRT[63]----------------- 120
Figure 5.10 Comparison of CIELUV L* and C* between 2 iPhones and CRT [64]--------------------------- 120
Figure 5.11 Comparison of the estimation results by subjects between CRT and iPhones[ 65]--122
Figure 5.12 Subjects’ estimation for colours on both CRT and iPhone5 [66]--------------------------------- 122
Figure 5.13 Colour checker depicted using four phones and CRT [67]---------------------------------------- 123
Figure 5.14 Comparison of CIECAM02 predictions with subject’s estimations [68]---------------------- 124
Figure 5.15 The prediction of iPhone lightness [69]------------------------------------------------------------ 125
Figure 5.16 iPhone lightness prediction for cyan samples [70]--------------------------------------------- 126
Figure 6.1 The prediction of iPhone colourfulness [71]------------------------------------------------------- 127
Figure 6.2 iPhone colourfulness prediction for cyan samples [72]------------------------------------------ 129
Elham Khodamoradi Page 11
Figure 6.3 The prediction of iPhone hue [73]----------------------------------------------------------------------- 130
Figure 6.4 iPhone hue prediction for cyan samples [74]------------------------------------------------------- 131
Figure 6.5 Mean Comparison lightness, colourfulness, hue for iPhone Experiment 1 to9 [75]------ 132
Figure 6.6 Plot between mean lightness (x) with lightness estimation by subject [76]----------------- 133
Figure 6.7 Comparison of iPhone (x-axis) with phones of LG-Nexus4 (top), Samsung (middle) and
Huawei (bottom) by modified CIECAM02 [77]------------------------------------------------------------------------ 134
Figure 6.8 Colour checker depicted using four telephones and CRT [78]---------------------------------- 136
Figure 8.1 Comparison of iPhone (x-axis) with phones of LG-Nexus4 (top), Samsung (middle) and Huawei (bottom) by modified CIECAM02. [80]----------------------------------------------------------------------- 140
Figure 8.2 Colour checker depicted using four phones and a CRT. [81]------------------------------------- 144
Elham Khodamoradi Page 12
List of tables
Table 2.1 Top 10 smartphone list as of May 2014 (Counter Point Research) [1] ___________________ 40
Table 2.2 Overview of the psychophysical experiments carried out in the study [2] _______________ 42
Table 2.3 Subject’s information [3] ______________________________________________________ 72
Table 3.1 CV values for Experiment 1 (Colour Sample) [4] ____________________________________ 74
Table 4.1 CV values for Experiment 2 (textured Colour Sample) [5] ____________________________ 84
Table 4.2 Summary of the CV values for Subjects A to G while performing Experiments 1 to 6 [6]__ 90
Table 4.3 Summary of the mean CV (%) values of subjects’ estimations for the 6 experiments [7]__ 93
Table 4.4 Values for iPhone Experiment [8] ________________________________________________ 95
Table 4.5 Summary of the mean CV (%) values of subjects’ estimations for the 6 experiments. [9]------96
Table 7.1 Values for iPhone Experiment. [10]-------------------------------------------------------------------------- 139
Elham Khodamoradi Page 13
Publication (Appendix D):
Evaluation of Colour Appearances Displaying on Smartphones, AIC2015, pp539-544,
2015, Tokyo, Japan.
Elham Khodamoradi Page 14
1. Introduction
A smartphone is a mobile telephone with more advanced computing capability
and connectivity than basic feature telephones, offering functionalities of typically
personal assistance, media player, digital camera, and a GPS navigation unit in
addition to the basic calling/receiving facilities. The global smartphone audience
had reached 67.0 million units in 2011 and was expected to reach 248.6 million
units by the end of 2015, [eMarkter, 2015]. The number of smartphone users
worldwide is expected surpass 2 billion in 2016, according to new figures. As
such, mobile sales are not only focusing heavily on smartphones, but also on the
more affordable option of feature phones that do not have an operating system.
As a result, mobile telephone penetration has surpassed 100% in many regions
of the world, including North America, Western Europe, Central and South
America, Central and Eastern Europe, and the Middle East [WeAreSocial, 2015].
One of the unexpected by-products of smartphones remains in the field of digital
photography. It appears that more photographs are taken using smartphones
than normal cameras. For example, among the few telephones used in this
study, the Apple iPhone 5 is the most popular 'camera' on Flickr.com [Flickr,
2015]. As a direct result, a large amount of money is being invested by mobile
developers to create photo apps in an attempt to satisfy the demands of 'serious'
camera phone photographers.
While using a smartphone to perform everyday activities, colour remains one of
the key factors, in particular in online shopping. Similar to any other digital
Elham Khodamoradi Page 15
device, a mobile telephone represents a digital image in a RGB colour space.
Therefore when an image is to be processed, it is usually first converted into the
colour space of, say, hue, lightness and colourfulness. In this way, the
dependency of RGB space on hardware devices can be circumvented, i.e. a
colour in one device usually does not appear nor measure the same as the one
in another device even with the same RGB values in both devices. This is
because the range of R, G, or B values are manually set to be the same (such as
[0, 255] for an 8-bit computer) for all devices regardless of their physical
measurements. Therefore in this research work, a number of smartphones are
studied with their colour appearances modelled using the CIE colour appearance
model, CIECAM02.
In this study, two phases of research work were conducted. Phase 1 was to train
observers to estimate colours using a magnitude estimation method and at the
same time to study a colour appearance model for modelling colour texture on
CRT monitors. Phase 2 is for mobile telephones. The structure phase one of the
thesis is as follows. Chapter 2 gives a comprehensive review of colour vision
theory, colour texture, colour spaces, colour models and the trend of
smartphones, which is then followed by Methodology in Chapter 3. Results and
data analysis are explained in Chapter 4, followed by Chapter 5 where the
conclusion is drawn up, and Chapter 6 where proposals for future work are
presented. The thesis concludes with References and Appendices.
Elham Khodamoradi Page 16
2. Literature Review
2.1 Overview of Colour Science
Colour is a visual experience generated primarily by three components, a visual
system (e.g. an eye), a coloured object, such as an apple, and a light that shines
on the object. Figure 2.1 illustrates the process of generating a colour. The light
source generates light illuminating an object. Some part the light is reflected from
the object to the eye and therefore the brain where the colour of the light is
observed [Hunt, 1992; Fraser, 2003; Sangwine, 1998; Davis, 1998; Berns, 2000].
Figure 2.1: Three essential components to generate a colour. [1]
As a part of the human vision system that involves a light source, an object, and
a processing system (e.g. brain), colour vision is the ability of the human eye to
separate the various frequencies of light waves (380~740nm) to allow us to
distinguish different colours through three colour cells, i.e. red, green and blue.
Accurate colour measurement requires reliable light sources, in order to make
objects react to the components of the spectrum, and therefore colour vision
Elham Khodamoradi Page 17
properties of human observers involved in the measurement process as shown
in Figure 2.2 [Neaves, 1998].
Figure 2.2: Three factors that are required to generate a colour. [2]
In 1666, Newton found that white light consists of a visible spectrum which
includes all the visible colours ranging from red, orange, yellow, green and blue
to violet [Luo, 1993a].
It was later found that colour is part of the electromagnetic spectrum with energy
in the range of 380nm to 780nm wavelength, the range that human vision can
see. Therefore the range between 380nm to 780nm is also known as ‘visible
light’ as illustrated in Figure 2.3 [Wyszecki, 1982].
Elham Khodamoradi Page 18
Figure 2.3. The Newton experiment. [3]
2.1.1 Light Source
The main light source is the Sun. In addition, there are a number of artificial light
sources including fluorescent lamps, or by heating up materials. There are two
ways to characterize a light source. One is to use a light spectral power
distribution (SPD). SPD is the amount of radiant power at each wavelength
represented by of the visible spectrum and denoted by P(). The other
common term to characterize light sources is colour temperature. Colour
temperature corresponds to the temperature of a heated blackbody radiator. The
colour of the blackbody radiator changes with the change of temperature. The
absolute temperature is measured using K standing for Kelvin.
For example, the blackbody radiator changes from black at 0K, to red at about
1000K, white at 4500K to bluish white at about 6500K. The colour temperature of
the Sun may vary during the time of the day (e.g. reddish at sunrise and bluish at
noon) and the weather conditions (e.g., sky with/without clouds) [Gevers, 2001].
Elham Khodamoradi Page 19
The Commission Internationale de L'éclairage (CIE) [CIE, 2015] has
recommended that average daylight has the colour temperature of 6500K and is
denoted by D65, whereas 5000K, or D50, represents overcast daylight. Figure
2.4 illustrates the relationship between colour temperature and spectral power
distribution [Kelly, 1963].
Figure 2.4: Relationship between colour temperature and spectral power distribution. [4]
2.1.2 Object
Coloured materials are called objects. The colour of an object is defined by the
reflectance or transmittance that is a function of wavelength. Reflectance is the
ratio of the light reflected from a sample to that reflected from a reference white
board, and is denoted by R() [Horne, 1998]
Usually, the colour reflected from or passed through an object is the product of
the SPD of the illuminant (P()) and the spectral reflectance of the object (R())
is computed by the formula [Berns, 2000; Berlin, 1969] shown in Eq. (2.1).
(2.1)
2.1.3 Observer
The observer measures light coming directly from a light source P () or light
which has been reflected (or transmitted) from objects in the scene R(). The
)()()( RPPR
Elham Khodamoradi Page 20
observer can be a colour camera or human eyes. For the human eye, the retina
contains two different types of light-sensitive receptors, called rods and cones.
Rods are more sensitive to monochromatic light and are responsible for vision in
twilight. Cones are responsible for colour perception and consist of three types of
receptors sensitive to Long (red), Middle (green) and Short (blue) wavelengths.
The response of these three cones to different wavelengths is shown in the
Figure 2.5 below [Color, 2010].
Figure 2.5: Response of three types of cone in relation to wavelength, where L, M, and S represent long (red), medium (green) and short (blue) wavelengths respectively. [5]
However, the sensation of a human observer cannot be measured by an
objective instrument. Therefore, experiments have to be conducted to measure
human observers’ spectral sensitivities to colours. The observers are asked to
match a test light, made of one single wavelength, by adjusting the energy levels
of three separate primary lights which are Red (700nm), Green (546.1nm), and
Blue (435.8nm) recommended by CIE. At each wavelength the amount of energy
was recorded for the three primary colours. The results of this matching are
called colour matching functions, usually denoted as )(x , )(y and )(z .
Also these colour matching functions can be treated as the colour response of
the eye [Hunt, 1995; Stitles, 2000] which are displayed in Figure 2.6. The
Elham Khodamoradi Page 21
tabulated numerical values of these functions are known collectively as the CIE
standard observer.
Figure 2.6: Human perception colour matching function [Fairchild M.D., Colour Appearance Models, 2nd Ed., Wiley, Chichester (2005)] [6]
2.2 Colour vision
2.2.1 Human Eye Structure
In the process of colour vision, the human eye is the last element in the system.
The retina is the light sensitive part of the eye. Its surface is coated with millions
of photoreceptors. These photoreceptors sense the light and pass electrical
signals through the optic nerve to stimulate the brain. There are two types of
photoreceptors, rods and cones [Neaves, 1998].
Figure 2.7: Human eye receptor. [7] [https://luminous-landscape.com/color-vision/]
Elham Khodamoradi Page 22
The human eye has a simple two element lens. The cornea is the front and the
lens is the back. The main function of the lens is to alter the image by changing
its shape; thinner for viewing far objects and thicker for near object. The cornea
and lens acting together form a small inverted image of the outside world on the
retina, the light-sensitive surface of the eye. [Neaves, 1998]
Figure 2.8: Anatomy of the Human Eye. [8] (www.studenthealth.ucla.edu/FormsDocuments/LayereoftheHumanEye)
2.2.2 Colour Vision Theory
Colour vision is the result of a system comprising the eye, the nervous system,
and the brain. Thomas Young [Young, 1802] first propounded the trichromatic
theory of colour vision including three types of cone receptors (or colour
receptors) in the eye, red, green, and violet, following Newton's earlier
investigation. In 1852 it was revised and elaborated by Helmholtz. The modified
theory is known as the Young-Helmholtz theory of colour vision [Helmhotz,
1924]. This assumed that the eye contained only three spectrally unique cone
Elham Khodamoradi Page 23
receptors, primarily red, green, and blue. In 1878 Edwald Hering provided
additional insight, proposing six independent colours, red, green, yellow, blue,
white, and black [Hering,1878]. These colours are registered by three opponent
colour systems, black-white, red-green, and yellow-blue. Thus an observer sees
colour in terms of redness or greenness, and yellowness or blueness. Both the
Young-Helmholtz theory and Hering theory paved the road for subsequent
research.
Since three types of independent receivers in the eye are required to match all
possible colours, normal colour vision is called trichromatic [Burnham, 1963].
The spectral response curves are used to describe the response of the eye at
different wavelengths. Curves plotting the amounts of R (red), G (green) and B
(blue) required to match a constant amount of power per small constant-width
wavelength interval at each wavelength of the spectrum for an observer are
called colour-matching functions and designated by symbols r(), g(), and b()
[Hunt, 1987].
The is the visible wavelength. These colour-matching functions were
determined independently by Guild [Guild, 1932] and Wright [Wright, 1928].
Figure 2.9 schematically shows the basic experimental arrangement [Billmeyer
1981]. The test colour produced by the test lamp is to be matched and displayed
in the bottom of the field of view. In the top, an observer sees an additive mixture
of beams of red, green, and blue lights. The composition of the lights is then
adjusted to match the test colour.
Elham Khodamoradi Page 24
Figure 2.9: A visual arrangement for producing a colour by mixing the light from three different coloured lamps [Billmeyer 1981]. [9]
In Guild's investigation [Guild, 1932], 7 observers made colour matches
throughout the visible spectrum. The amounts of red, green, and blue were
obtained and were expressed in terms of Guild's instrumental stimuli
(heterochromatic primaries obtained with coloured filters) after the units have
been adjusted to give equal amounts of the primaries to match the National
Physical Laboratory standard white at the National Physical Laboratory (NPL) at
Teddington, U.K .
On the other hand, Wright [Wright, 1928], utilised monochromatic primaries at
650, 530, and 460nm. Their units (the quantity of each primary) were adjusted so
that equal amounts of red and green stimuli were required in a match of a
monochromatic yellow (582.5nm), and equal amounts of green and blue were
required in a match of a monochromatic cyan (494nm). Using 10 observers,
Wright carried out the experiment at Imperial College, London, England.
Elham Khodamoradi Page 25
Although remarkably different techniques were applied by the two researchers,
their results could be converted to the same set of primaries due to the algebraic
nature of colour. The primaries chosen were R (700nm), G (546.1nm) and B
(435.8nm). The units of R, G, and B were adjusted to be equal in a match on an
'equal-energy' white (a white in which the energy per unit wavelength was
constant through the visual spectrum). The amounts of each primary used to
obtain a match are known as tristimulus values, R, G, and B. Tristimulus values
can be converted into chromaticity coordinates as given below.
𝑟 =𝑅
𝑅+𝐺+𝐵, 𝑔 =
𝐺
𝑅+𝐺+𝐵, 𝑏 =
𝐵
𝑅+𝐺+𝐵 (2.2)
2.2.3 Colourimetry
As mentioned in Section 2.1, colour perception requires three factors: a source
of light, an object and a detector, usually the eye and brain. Each of these can be
described by an appropriate curve across the visible wavelength. The
combination of these comprises a colour or colour stimulus. A quantitative
method to describe a colour stimulus is shown in Figure 2.10 [Billmeyer, 1981].
Figure 2.10: Quantitative definition of a stimulus [Billmeyer, 1981]. [10]
Elham Khodamoradi Page 26
In 1931, the International Commission on Illumination (CIE) [CIE, 2007] adopted
a system of colour specification which has lasted to the present time, known as
the CIE system of colorimetry (see Section 2.3). In this system, a colour is
defined by a set of X, Y, Z values, called tristimulus values. Two samples of
identical material should be judged as an exact match when their tristimulus
values are the same.
2.2.4 Subject Estimation
Colour can also be subjectively specified by means of visual percept. In the OSA
Colour Committee terminology the word "colour' is clearly defined as follows:
"Colour consists of the characteristics of light other than spatial and temporal
inhomogeneities; light being the aspect of radiant energy of which a human
being is aware through the visual sensations which arise from the stimulation of
the retina of the eye". Colour appearance is defined by Judd as "the colour
perceived to belong to the visual object to which attention is directed" [Judd,
1965]. The hue, colourfulness and lightness, abstracted from complete visual
experiences, are used to represent dimensions along which colour may vary
independently. Hue is defined as the attribute of a visual sensation according to
which an area appears to be similar to one, or to proportions of two, of the
perceived colours, red, yellow, green, and blue. Colourfulness is the attribute of
a visual sensation according to which an area appears to exhibit more or less of
its hue. Chroma is the colourfulness of an area judged in proportion to the
brightness of a similarly illuminated area that appears to be white or highly
transmitting. Saturation refers to the colourfulness of an area judged in
proportion to its brightness. Brightness implies the attribute of a visual sensation
Elham Khodamoradi Page 27
according to which an area appears to exhibit more or less light. Lightness is
the brightness of an area judged relative to the brightness of a similarly
illuminated area that appears to be white or highly transmitting [CIE, 1970;
Evans, 1964].
There are two colour modes depending upon the colour being perceived. One is
'object mode' and another is 'aperture' (or 'light-source mode'). Object mode is
when a visual object appears as being illuminated by an external emitting light.
This may be observed when the object is viewed under the surrounding of the
other objects. Aperture colour implies that the visual object is emitting light by
itself. A colour is perceived as a hole filled with a colour light when the
surrounding field of the visual object is completely dark. Colours seen in these
special circumstances are often referred as 'unrelated colours'. An aperture
colour may also be observed in its background to other visual objects which,
however, are usually of a low luminance. The above two modes of colour,
however, cannot be perceived simultaneously [Wyszecki, 1973].
2.3 Colour Space and Model
Although it is possible to represent a colour by using a spectrum of wavelengths,
it is not convenient to represent a colour using so many parameters across
visible wavelengths. Therefore in 1932, the very first mathematically defined
colour space, CIEXYZ, was recommended [CIE, 1932].
Since then the CIE (Commission Internationale de l'Éclairage) has
recommended several other standard colour spaces to represent a colour,
including CIELUV [Colour Spaces, 1932], and CIELAB, and one colour
Elham Khodamoradi Page 28
appearance model, CIECAM02 [Moroney, 2002]. A colour space is a
mathematical representation, which is a three-dimensional orthogonal coordinate
system.
2.3.1 CIEXYZ
CIEXYZ was the first mathematical model recommended in 1931 by the
Commission Internationale de l'Éclairage (CIE) to represent a colour, and is also
called the CIE 1931 XYZ colour space.
The CIE XYZ colour space was derived from a series of experiments done in the
late 1920s by W. David Wright and John Guild [Wright, 1928]. If a colour is given
by I(), and )(x , )(y and )(z are matching functions, Eq. (2.3) gives the
values of X, Y, and Z, which are also called tristimulus values for the colour,
whereas is the wavelength measured in nanometres
𝑋 = ∑ 𝐼(𝜆)�̅�(𝜆)
𝑌 = ∑ 𝐼(𝜆)�̅�(𝜆) (2.3)
𝑍 = ∑ 𝐼(𝜆)𝑧̅(𝜆)
2.3.2 CIE xy chromaticity diagram
For each colour, three tristimulus values X, Y, and Z can be used to represent it.
Therefore a full plot of all the visible colours is a 3-dimensional figure. On the
other hand, in terms of colour perception, a colour can be divided into two parts:
brightness and chromaticity, the colour content. For example, the colour white is
Elham Khodamoradi Page 29
a bright colour, while the colour grey is considered to be a less bright version of
that same white.
In other words, the chromaticity of white and grey are the same while their
brightness differs. The CIE XYZ colour space is therefore deliberately designed
so that the Y parameter is a measure of the brightness or luminance of a colour.
The chromaticity of a colour is then specified by the two derived parameters x
and y, two of the three normalized values which are functions of all three
tristimulus values X, Y, and Z as calculated in Eq. (2.4).
(2.4)
The plot of y against x for all the colours concerned is then called the xy
chromaticity diagram as shown in Figure 2.11.
Elham Khodamoradi Page 30
Figure 2.11: The xy chromaticity diagram for all the visible colours [http://wolfcrow.com/blog/what-is-color-space-and-the-tristimulus-values-xyz] [11]
2.3.3 CIELUV
In 1976, the CIE defined a new colour space CIELUV to enable us to get more
uniform and accurate models [CIE, 2009]. Sometimes, it is also called the
universal colour space. This colour representation is derived from CIEXYZ. The
CIE LUV colour space is a perpetually uniform derivation of a standard CIEXYZ
space [Hunt, 1998].
As the distribution of the colours in xy chromaticity coordinates is not uniform, a
new colour chromaticity coordinate, u’v’, was recommended by CIE in 1976 [CIE,
2009], can be calculated using Eq. (2.5).
ZYX
Yv
ZYX
Xu
315
9'
315
4'
(2.5)
Elham Khodamoradi Page 31
(a) (b)
Figure 2.12: Chromaticity colour space xy (a) and u'v'(b). [http://dba.med.sc.edu/price/irf/Adobe_tg/models/ciexyz.html] [12]
Chromaticity diagrams only show proportions of tristimulus values, and not their
actual magnitudes. They are only strictly applicable to those colours with the
same luminance.
Colours however, differ in both chromaticity and luminance, and some methods
of combining these variables are required. In 1976, the CIE used the CIELUV
colour space as the perceptually uniform colour spaces whose expressions are
defined below [CIE, 2008].
Elham Khodamoradi Page 32
16)(116*0
Y
YfL
, if Y/Y0 >0.008856, else )(3.903*
0Y
YL
)''(*13* 0uuLu
)''(*13* 0vvLv
*)/*arctan(uvH
22 *)(*)( vuC
(2.6)
where 0'u , 0'v are the values of u’, v’ for the appropriately chosen reference
white. The L component has the range of [0,100].
2.3.4 CIELAB
Another colour space recommended by CIE is CIELAB [CIE, 2008]. A lab colour
space is a colour-opponent space with dimension (L) for lightness (a) and (b) for
the colour-opponent dimensions, based on nonlinearly compressed CIE XYZ
colour space coordinates. Eq. (2.7) gives the calculation of CIELAB.
(2.7)
where
(2.8)
The Chroma and hue are calculated as Eqs. (2.9) and (2.10).
(2.9)
H = arctan (b*/a*) (2.10)
Elham Khodamoradi Page 33
CIELAB chromatic adaptation is compatible for near-daylight illuminants as well
as medium grey background and surround and moderate luminance levels. The
only weakness in CIELAB is that it cannot account for changes in background
surround luminance cognition therefore it is unable to predict brightness and
colourfulness. To overcome this limitation CIELAB can be replaced with a more
accurate model such as CAT02.
While content-based image retrieval (CBIR) has been researched for nearly
three decades, it remains debatable on whether it can ever meet users’
expectations. On the one hand, due to the exponential increase of digital images,
the demand for retrieving relevant data in an efficient, sufficient and effective way
is high, especially amongst those images on the Internet that are not properly
labelled, to complement the current text-based approaches. On the other, the
subjectiveness of the interpretation of an image can lead to the association
varying considerably between visual contents such as colour, texture and shape.
To strengthen this gap, in general, the development of algorithms for extraction
of those visual features has been endeavoured to endorse with human vision
theories by considering colour textured samples. More importantly, colour, in
many ways, among many visual features presented in an image, tends to be a
decisive factor, such as in determining the purchase of clothes or the selection of
wall papers. Additionally, with regard to internet retrieval, colour can be
processed faster in real time due to the availability of many well established
colour theories and models, which has led to much wider employment of colour
spaces or models in content-based image retrieval (CBIR) systems than any
other feature models. While colour-based approaches employ colour spaces of
Elham Khodamoradi Page 34
mainly RGB, HSV, CIELAB and CIELUV, those spaces do not take simultaneous
texture into consideration.
Although an image begins by being represented using RGB colour space when it
is in a digital form, it is usually converted into HSI (hue, saturation and intensity)
colour space to circumvent the dependency nature of RGB space on hardware
devices, i.e. a colour in one device usually does not appear nor measure the
same as one in another device even with the same RGB values in both devices.
This is because the range of R, G, or B values are manually set to be the same
(such as [0, 255] for an 8-bit computer) for all monitors regardless of their
physical measurements. On the other hand, HSI space agrees more with human
vision theories. To further improve the fitness between users’ perception and
retrieved results, a colour appearance model has been proposed earlier in this
section. In this study attention is paid to the development of distance formulae,
with textured colour as the focus.
As mentioned above, the colour appearance model usually studied in a centre-
surround 3-field paradigm consists of background, induction field and a test
colour. Hence, this research comprises two parts with Phase 1 establishing a
colour appearance model that in turn is employed for colour textured samples in
order to establish the effect of texture in colour estimation by an observer,
carried out in Phase 2. The same experiment was carried out on mobile
telephones to evaluate their colour appearance by employing the CIECAM02
colour appearance model.
2.3.5 The Input and Output Information for CIECAM02
With regard to the representation of the colour appearance of an image, in this
investigation, the perceptual colour attributes of lightness (J), colourfulness (M)
Elham Khodamoradi Page 35
and hue (H) are employed, which are detailed in Appendix B and briefly listed
below.
Given a set of tristimulus values in XYZ, the corresponding LMS values can be
determined by the MCAT02 transformation matrix:
Once in LMS, the white point can be adapted to the desired degree by choosing
the parameter D. For the general CAT02, the corresponding colour in the
reference illuminant is:
where the Yw / Ywr factor accounts for the two illuminants having the same
chromaticity but different reference whites. The subscripts indicate the cone
response for white under the test (w) and reference illuminant (wr). The degree
of adaptation (discounting) D can be set to zero for no adaptation (stimulus is
considered self-luminous) and unity for complete adaptation (colour constancy).
Elham Khodamoradi Page 36
Figure 2.13: Colour constancy
[Moroney, Nathan; Fairchild, Mark, 2002] [13]
In practice, it ranges from 0.65 to 1.0, as can be seen from the diagram.
Intermediate values can be calculated by:
where surround F is as defined above and LA is the adapting field luminance in
cd/m.
Figure 2.14: Post-adaptation
[Moroney, 2002] [14]
Elham Khodamoradi Page 37
In CIECAM02, the reference illuminant has equal energy (Lwr = Mwr = Swr = 100)
and the reference white is the perfect reflecting diffuser (i.e. unity reflectance,
and Ywr = 100) hence:
Furthermore, if the reference white in both illuminants have the Y tristimulus
value (Ywr = Yw)
After adaptation, the cone responses are converted to the Hunt–Pointer–Estévez
space by going to XYZ and back:
Elham Khodamoradi Page 38
Figure 2.15: Log L′a vs. log L′ for LA = 200 (FL = 1)
[Moroney, 2002] [15]
Finally, the response is compressed based on the generalized Michaelis–Menten
equation [Moroney, 2002]
where FL is the luminance level adaptation factor.
As previously mentioned, if the luminance level of the background is unknown, it
can be estimated from the absolute luminance of the white point as LA = LW / 5
Elham Khodamoradi Page 39
using the "medium gray" assumption [Schanda, 2007] (The expression for FL is
given in terms of 5LA for convenience.) In photopic conditions, the luminance
level adaptation factor (FL) is proportional to the cube root of the luminance of the
adapting field (LA). In scotopic conditions, it is proportional to LA (meaning no
luminance level adaptation). The photopic threshold is roughly LW = 1 (see
Figure 2.15).
CIECAM02 defines correlates for yellow-blue, red-green, brightness, and
colorfulness. Let us make some preliminary definitions.
The correlate for red–green (a) is the magnitude of the departure of C1 from the
criterion for unique yellow (C1 = C2 / 11), and the correlate for yellow–blue (b)
is based on the mean of the magnitude of the departures of C1 from unique red
(C1 = C2) and unique green (C1 = C3).
The 4.5 factor accounts for the fact that there are fewer cones at shorter
wavelengths (the eye is less sensitive to blue). The order of the terms is such
that b is positive for yellowish colors (rather than blueish).
The hue angle (h) can be found by converting the rectangular coordinate (a, b)
into polar coordinates:
Elham Khodamoradi Page 40
To calculate the eccentricity (et) and hue composition (H), determine which
quadrant the hue is in with the aid of the following table. Choose i such that hi ≤
h′ < hi+1, where h′ = h if h > h1 and h′ = h + 360° otherwise.
Red Yellow Green Blue Red
i 1 2 3 4 5
hi 20.14 90.00 164.25 237.53 380.14
ei 0.8 0.7 1.0 1.2 0.8
Hi 0.0 100.0 200.0 300.0 400.0
Table 2.1 Hue angel chart
[Schanda, 2007]
Calculate the achromatic response A:
where:
The correlate of lightness is:
Elham Khodamoradi Page 41
where c is the impact of surround (see above), and:
The correlate of brightness is:
Then calculate a temporary quantity t:
The correlate of chroma is:
The correlate of colorfulness is:
The correlate of saturation is:
2.3.6 CIECAM
To model a colour appearance, CIE (Commission Internationale de l'éclairage)
has recommended a colour appearance model, CIECAM02 [Moroney, 2002],
Stemming from Hunt’s early colour vision model [Luo 1993a, Luo 1993b, Luo
1993c] employing a simplified theory of colour vision for chromatic adaptation
together with a uniformed colour space, CIECAM02 can predict the change of
Elham Khodamoradi Page 42
colour appearance as accurately as an average observer under a number of
given viewing conditions, in which the way that the model describes a colour is
reminiscent of subjective psychophysical terms, i.e., hue, colourfulness, Chroma,
brightness and lightness.
To start with, the model takes into account of the tristimulus values (X, Y, and Z)
of the stimulus, its background, its surround, the adapting stimulus, the
luminance level, and other factors such as cognitive discounting of the illuminant.
The output of the colour appearance model includes mathematical correlates for
perceptual attributes. Table 2.2 summarises the input and output parameters
from CIECAM02.
TABLE 2.2. THE INPUT AND OUTPUT INFORMATION FOR CIECAM02.
Input Output
X,Y, Z: Relative tristimulus values of colour stimulus
XW, YW, ZW: Relative tristimulus values of white
La: Luminance of the adapting field (cd/m*m) = 1/5
of adapted D65;
Yb: Relative luminance of the background = 0.2;
Surround parameters: c, Nc, F = 0.41, 0.8, 0.9
respectively for luminous colours (i.e. monitor).
Lightness (J)
Colourfulness (M)
Chroma (C)
Hue angle (h)
Brightness (Q)
Saturation (S)
With regard to the representation of the colour appearance of an image, in this
investigation, the perceptual colour attributes of lightness (J), Chroma (C),
Elham Khodamoradi Page 43
colourfulness (M) and hue angle (h) are employed, which are calculated in Eqs.
2.11 to 2.14 respectively.
𝐽 = 100(𝐴
𝐴𝑤)𝑐𝑧 (2.11)
𝐶 = 𝑡0.9(𝐽
100)0.5(1.64 − 0.29𝑛)0.73 (2.12)
𝑀 = 𝐶𝐹𝐿1/4 (2.13)
ℎ = 𝑡𝑎𝑛−1(𝑏
𝑎) (2.14)
where:
𝐴 = [2𝑅𝑎′ + 𝐺𝑎
′ +1
20𝐵𝑎
′ − 0.305] 𝑁𝑏𝑏 (2.15)
𝑡 =50(𝑎2+𝑏2)
12100𝑒(
10
13)𝑁𝑐𝑁𝑐𝑏
𝑅𝑎′ +𝐺𝑎
′ +21
20𝐵𝑎
′ (2.16)
𝑎 = 𝑅𝑎′ −
12𝐺𝑎′
11+
𝐵𝑎′
11 (2.17)
𝑏 =1
9(𝑅𝑎
′ + 𝐺𝑎′ − 2𝐵𝑎
′ ) (2.18)
and ''' ,, aaa BGR indicate the post-adaptation cone responses with detailed
calculations specified in [Wu, 2007] whereas WA refers to the A value for
reference white. Constants cbbb NN , are calculated as:
Elham Khodamoradi Page 44
𝑁𝑏𝑏 = 𝑁𝑐𝑏 = 0.725(1
𝑛)0.2 (2.19)
where n = 𝑌𝑏/𝑌𝑤 , with Yb and YW representing the Y value for both background
and reference white respectively. The work flow of CIECAM02 is illustrated in
Figure 2.16 with its full formula given in Appendix A.
Figure 2.16: The work flow of CIE colour appearance models. [www.researchgate.net/publication/227991182] [16]
2.3.7 Colour Appearance Datasets
Colour appearance models based on colour vision theories has been developed
to fit various experimental data sets, which were carefully produced to study
particular colour appearance phenomena. Over the years, a number of
experimental data sets were gathered to test and develop various colour
appearance models. Luo [Luo, 1995] investigated a data set of saturation
correlates using the magnitude estimation method. The data accumulated played
an important role in the evaluation of the performance of different colour
appearance models and the development of CIECAM97s and CIECAM02. In this
Elham Khodamoradi Page 45
study, the experimental design is similar to Luo’s [Luo, 1995] in order to be
comparable.
2.4 Colour Appearance Phenomena
Colour perceived presentation is influenced by various colour appearance
phenomena caused by any undetected change in viewing conditions in the
colour reproduction process. Therefore to have a better understanding of colour
research it is essential to analyse colour appearance phenomena to represent
the colour appearance qualitatively and quantitatively and accurate colour
reproduction in the easiest way possible. In this research, colour appearance
and common colour appearance phenomena were analysed. Also the basic
theory of colour appearance in colour reproduction was studied.
The important colour appearance phenomena were studied in this research
based on the viewing conditions and the interaction of the colours. The results
help to describe the colour appearance models significantly and accurately.
Almost all colour-appearance phenomena describe relationships between
changes in viewing conditions and changes in appearance. If two stimuli do not
match in colour appearance when (XYZ) 1 = (XYZ) 2, then some aspect of the
viewing conditions is different.
Colour appearance phenomenon is that the perceived colour changes with
changes in the viewing conditions. Even under the same viewing conditions, the
colour appearance perceived is not always the same, as the perceived colour
stimuli are influenced by the neighbourhood pixels [Qin-ling Dai, 2012].
Elham Khodamoradi Page 46
Figure 2.17: Simultaneous Contrast/ Induction Fig 2.18: Influence of simultaneous contrast on colour. [17,18]
In Figure 2.17, the colour perception is inconsistent if the background is
changed. As shown in Figure 2.18, with the dark background, the colour patch
seems lighter but with a beige background it seems darker.
In Figure 2.19, two colour patches with the same chromatic colour value which
lie on a dark background appear lighter and brighter than that sets where the
background is of whiter shade.
Figure 2.19: Coloured object set in pink Figure 2.20: Object set in yellow and and white background. [19] blue background. [20]
Simultaneous contrast or colour induction causes the colour vision implication of
perception which causes the whole colour appearance to change by changing
the background colour stimulus. It clearly shows that the colour stimulus will be
darker on a lighter background while appearing darker on a lighter background.
Red-green and yellow-blue will cause the corresponding induction colour vision,
Elham Khodamoradi Page 47
i.e. red will induct green, green will induct red, yellow will induct blue and blue will
induct yellow.
2.4.1 Crispening Effect
The Crispening effect [Moroney, 2002] refers to the apparent increase in
perceived magnitude of colour differences when the background is similar in
colour to the object.
As shown below in Figure 2.21, the two blue stimuli give the impression of
having greater lightness variance on the pale blue background than on either the
white or dark blue backgrounds. [Moroney, 2002].
Figure 2.21: The pairs of blue patches are physically identical on all three backgrounds. [21]
2.4.2 Hunt Effect. (Colourfulness increases with luminance)
In general, colourfulness increases with luminance, this is known as the Hunt
effect [Moroney, 2002]. It underlines the effect of absolute lightness on colour
appearance. The Hunt effect is a common phenomenon in cross-media colour
Elham Khodamoradi Page 48
reproduction. However it is out of reflection in foundation colorimetry and should
be predicted in colour appearance models. It has a very significant impact on
CIECAM02, but because of its complexity the Hunt model itself is difficult to use
[Moroney, 2002].
Figure 2.22: Hunt effect sample [http://facweb.cs.depaul.edu/sgrais/color_context] [22]
2.4.3 Stevens Effect (contrast increases with luminance)
The Stevens’s effect is that lightness relation increases with luminance of the
adaptive field. It can be demonstrated in the figure below. The dark tone will be
darker while the light tone will be lighter with the luminance increases, i.e. the
contrast increases [Moroney, 2002].
Elham Khodamoradi Page 49
Figure 2.23: Stevens effect [http://facweb.cs.depaul.edu/sgrais/color_context] [23]
2.4.4 Bezold Brucke Effect
Colorimetric purity of monochromatic light will change when it mixes with white
light. The colour tone will change for monochromatic light with different lightness.
According to Bezold Brucke, colour tune drift effects hue change with luminance.
The phenomenon can be illustrated as in Figure 2.24 below. [Moroney, 2002].
Figure 2.24 : Bezold Brucke effect [http://facweb.cs.depaul.edu/sgrais/color_context] [24]
Elham Khodamoradi Page 50
In colour reproduction and perception, there are also other colour appearance
phenomena because of the changing viewing conditions, however, the above are
the most common factors.
2.4.5 Colour Constancy
Colour Constancy is the ability to distinguish colours of objects, invariant to the
colour of the light source, in other words, constancy of observed colour relations
under dissimilar illuminants. The resulting changes in the spectrum of the light
reflected from a scene are readily apparent over the course of a day [Romero,
Hernández-Andrés, Nieves, & García, 2003], with the gamut of colours at sunset
almost doubling under the mixture of direct and indirect illuminations [Hubel,D,
1995].
The human visual system is extremely capable of recognising the changes in
both the level and the colour of the lighting which result in this capability known
as adaptation; therefore objects tend to be recognized as having nearly the same
colour in very many conditions. This human vision recognition system is well
known for its colour constancy. Having said that, colour constancy is only
estimated, and considerable changes in colour appearance can sometimes
occur, as in the tendency for colours that appear navy in daylight to appear
distinctly bluish in tungsten light; but still colour constancy is an extremely
powerful and important effect in colour perception.
Elham Khodamoradi Page 51
Figure 2.25. Three optical illusions demonstrating colour constancy in action. A. The checkerboard illusion of Edward Adelson. B. The cube illusion of R. Beau Lotto. C. The cross-piece illusion of R. Beau Lotto [Dimension of colour by Dr David. J.C. Briggs [Briggs, 1998]. [25]
2.4.6 Metamerism
Metamerism is the phenomenon wherein two coloured samples will seem to be
of the same shade under one light source but will appear to have different
shades under a second source. The two different colours have identical colour
appearance under one light source (metameric match) but look different under
another light source (metameric mismatch). Metamerism is the dissimilarity of the
reflection curves of two colours which look identical under a given form of lighting
(e.g. Florescent light, electric bulb light, daylight).
Figure 2.26: When two colours having different spectral compositions match one another, they are said to be metameric, or metamers, and the phenomenon is called metamerism. [http://dba.med.sc.edu/price/irf/Adobe_tg/color/variables.html]. [26]
The CIE has recommended that the degree of metamerism for changes of
illuminant can be allowed for and can be found by calculating an Illuminant
Elham Khodamoradi Page 52
Metamerism Index, M, consisting of the size of the colour difference between a
metameric pair caused by substituting, in place of a reference illuminant, which
in this case is the test illuminant having a different spectral composition. The
preferred reference illuminant is known as the Standard illuminant D65 for
daylight illumination. [Hunt, 1998].
Colour perception is different when the same colour is viewed under different
conditions. Mostly the viewing conditions include lighting condition, medium,
background, surround and the observer.
The reason for conducting colour experiments under controlled viewing
conditions is to have the most accurate result by overcoming the issues that
have been raised. Associated with the specification of viewing fields for colour
appearance models are the various definitions of standard viewing conditions
used in different situations. This attempts to minimize difficulties with colour
appearance by outlining appropriate viewing field configurations [Moroney, 2002].
2.5 Chromatic-adaptation models
Chromatic-adaptation models offer nominal scales for colour appearance. Two
stimuli in their relative viewing conditions match each other but might be more or
less different in their colour properties. To verify the colour difference and have
more accurate results, we need colour-appearance models to evaluate and
measure: Lightness, Brightness, Hue, Chroma, and Colourfulness. The Colour
Appearance Model provides mathematical formulae to transform physical
Elham Khodamoradi Page 53
measurements of the stimulus and viewing environment into different attributes
of colour, e.g. lightness, colourfulness and hue.
In the vision subject area, three types of adaptation are considered important
issues, namely light, dark, and chromatic.
Light adaptation tells us that increases in the overall level of illumination has a
negative effect on visual sensitivity. In other words, increasing the illumination
level will decrease visual sensitivity [Moroney, 2002].
Dark adaptation is similar to light adaptation, as dark adaptation is the increase
in visual sensitivity experienced upon decreases in luminance level.
Time-course is one of the important features of chromatic adaptation
mechanisms. This issue in regard of colour appearance judgements has been
explored in detail [Fairchild, 1992], [Fairchild, 1995]. The outcome of these
studies suggests that the sensory mechanisms of chromatic adaptation are
about 90% complete after 60 seconds for changes in adapting chromaticity at
constant luminance. As a result, the minimum time for making critical judgments
for an observer have been suggested as around 60 seconds. Adaptation is
slightly slower when significant luminance changes are involved [Hunt, 1952].
Cognitive mechanisms of adaptation rely on knowledge and interpretation of the
stimulus configuration. Having said that, in some unusual viewing situations, the
time required to interpret the scene can be substantially longer than mentioned
above.
Elham Khodamoradi Page 54
2.5.1 Colorimetry
Colorimetry is the branch of colour science dealing with specifying statistically
the colour of a physically defined visual stimulus by the observer. In particular, it
is concerned with:
1) when an object is viewed by an observer with normal colour vision, under the
same observing conditions, readings and measurements of the same object
should appear similar.
2) stimuli that appear similar should have the same specification.
3) the numbers comprising the specification are functions of the physical
parameters defining the spectral radiant power distribution of the stimulus
Trichromatic generalization (R,G,B).
2.5.2 Colour-matching functions
Given a monochromatic stimulus Qλ wavelength, it can be written as
Qλ= RλR + GλG + BλB
where Rλ, Gλ, and Bλ are the spectral tristimulus values of Qλ .
Assume an equal-energy stimulus Eλ whose mono-chromatic constituents are
Eλ (equal-energy means Eλ ≡ 1).
The equation for a colour match involving a mono-chromatic constituent Eλ of E
is. Eλ= rλR + gλG + bλB,
Elham Khodamoradi Page 55
where rλ, gλ, and bλ, are the spectral tristimulus values of Eλ.
The sets of such values are called colour-matching functions
Figure 2.27: Colour-matching Diagram. [Harris, 1990] [27]
2.5.3 Chromaticity Diagrams
We can normalize the colour-matching functions and thus obtain new quantities.
r (λ) = r (λ) / [r (λ) + g(λ) + b(λ)]
g(λ) = g(λ) / [r (λ) + g(λ) + b(λ)]
b(λ) = b(λ) / [r (λ) + g(λ) + b(λ)]
with r(λ) + g(λ) + b(λ) = 1
The locus of chromaticity points for monochromatic colours so determined is
called the spectrum locus in the (r, g) - chromaticity diagram.
Elham Khodamoradi Page 56
Figure 2.28: Chromaticity diagrams. [Hunt, (2011). "The Colour Triangle"]. [28]
2.6 Experimental Method of Magnitude Estimation to Determine Colour Appearance
In this research, the magnitude estimation method is applied to estimate the
appearance of each colour. Each observer is asked to make a subjective
estimate of the magnitude of visual attributes. The attributes might be lightness,
brightness, colourfulness, saturation, chroma, and hue. The observer simply
assigns a number that in his or her view corresponds to the magnitude of the
chosen attribute in the sample being viewed. Alternatively, the observer might be
asked to make a subjective estimate of the attribute on some more clearly
defined scale, usually an equal-interval scale, or to compare two samples for an
estimating parameter [Ishak,1970; Stevens, 1957].
Elham Khodamoradi Page 57
The magnitude estimation technique was first tested by Stevens et al.
[Stevens,1957] and has recently gained in general acceptance. It is a subjective
scaling technique by which the magnitudes of perceived attributes are scaled.
Rowe [Rowe,1973] and Padgham [Padgham,1973] carried out their work to
scale hue and saturation. They concluded that a surprising degree of precision
can be achieved using this technique. In Ishak et al's study [Ishak, 1970], two
observers made estimations in terms of hue, saturation and lightness for 60
surface colours on seven backgrounds (Black, Grey, White, Red, Yellow, Green,
and Blue background). They compared their results to those of Helson et al.
[Helson,1952] (using the memory method), and Wassef, Hunt and Gibson
[Gibson, 1967] (using the binocular matching method). The results showed that
the magnitude estimation method was reliable in producing results similar to
those found using other methods. They concluded that the method was suitable
for measuring colour appearance under a variety of viewing conditions. Following
their study, Nayatani et al. [Nayatani,1972] examined the precision of this
method between and within observers, and reconfirmed its effectiveness. They
made assessments for three attributes of 100 object colours by a panel of fifteen
observers. A fluorescent lamp with a high colour-rendering index was used.
Results showed a good agreement with those obtained by Ishak et al. This
method was later employed by Bartleson [Bartleson,1979], Pointer [Pointer,
1980], and Luo et al [Luo 1991a, Luo 1991b].
In using a magnitude estimation technique, an observer simply views the test
sample and assigns numbers or names that correspond to the colour attributes
of its subjective appearance. Normally they are lightness, brightness, saturation,
colourfulness, and hue.
Elham Khodamoradi Page 58
Lightness is a subjective attribute that has been studied thoroughly by Stevens et
al. and by many others [Stevens 1957; Luo 1991a, Luo 1991b]. As far as the
method applied to reflecting surfaces, it was mainly grey content that was
examined [Stevens, 1963].
Brightness is defined by the CIE as the attribute of a visual sensation according
to which an area appears to exhibit more or less light [Bartleson,1980]. It is a
perceptually absolute quantity and has an absolute zero modulus without upper
limit. For many years attempts have been made to characterise perceived
brightness as a function of stimulus luminance. A variety of predictive equations
has been proposed. Stevens et al. [Stevens, 1963] specified brightness as a
power function of luminance. Bartleson's brightness-scaling experiments with a
complex stimulus field showed that the resulting brightness vs luminance
functions are not simple power functions but are nonlinear in log-log coordinates
[Indow, 1966].
For estimating hue, four to six names of basic or unique colours are commonly
used among which are Red, Yellow, Green, Blue and the two intermediate hues
orange and yellowish-green. For colour appearance between the unique colours
interpolations are used either in numerical form [Indow,1966] such as "80%
green, 20% yellow", or as combination names such as Blue-Green. This method
is closely associated with the NCS Colour System.
Earlier magnitude estimation experiments were conducted using saturation
rather than colourfulness. Saturation assessments were reported by many
researchers [Indow,1966]. In these studies, observers were asked to scale the
saturation of a test colour on a scale which had fixed points at both ends. One
Elham Khodamoradi Page 59
end (zero) represented a colour with no saturation (a neutral colour), and the
other end (100) represented the most saturated colour that the observer could
imagine having the same hue as the test colour. The test colour was then scaled
as a number between these two end points. This led to difficulties in analysing
the data because the most saturated colour varied in absolute saturation for
different hues, for example, a most saturated blue could be more saturated than
a most saturated yellow [Pointer,1980].
The concept of colourfulness was introduced by Hunt [Hunt, 1952] to denote the
attribute of a visual sensation according to which an area appears to exhibit
more or less chromatic colour. Pointer's [Pointer,1978] results showed that this
concept is meaningful to the observers who were asked to rank colour chips in
order of colourfulness and also able to scale the colourfulness of each individual
chip. In his experiment, colourfulness was scaled under various luminance levels
and backgrounds. A correlation coefficient of 0.97 was obtained between the
mean of saturation and colourfulness. This suggested that there was a high
degree of correlation between these two attributes. He concluded that
colourfulness was a useful concept which observers were well able to scale, and
may be more easily scaled than saturation or Chroma. If a full measure of the
appearance of a colour is required, colourfulness can provide changes in
chromatic response caused by the luminance levels.
Elham Khodamoradi Page 60
2.7 Summary
In summary, a magnitude estimation method as described below was used to
scale each colour in terms of lightness, colourfulness, and hue. In other words,
lightness represents the degree of brightness a colour is showing. With the
reference white being 100 and an imaginary black being zero, observers are
asked to give an estimation between 0 and 100 that is in proportion to the
reference white. For example, if a test colour appeared half as bright as the
reference white to a subject, 50% of lightness was assigned to the colour.
On the other hand, colourfulness shows the amount of hue appearing in a test
colour, which is in proportion to the reference colourfulness that is given
colourfulness of 40, e.g. if a colour has a colourfulness of 60, this colour should
display 1.5 times as colourful as the reference colourfulness colour. Neutral
colours, including white, black, and grey colours have zero colourfulness. For
hue estimation, four opponent hue scales were employed, which were red
against green and yellow against blue. The hue of a test colour was then
estimated with reference to any two neighbouring colours. For example, a hue
can appear 80% yellow and 20% red or 50% green and 50% blue. A colour
cannot show the contents of opponent hues, i.e. an answer of ‘30% green and
70% red’ is invalid (red and green are opponent hues), whilst neutral colours
have no content of hue at all. This hue scale was later converted into a hue
angle of 0 to 400 with 0 (400) being red, 100 yellow, 200 green and 300 blue for
data analysis.
Elham Khodamoradi Page 61
Furthermore, all observers’ vision was tested before conducting the experiment.
Figure 2.29 shows an example of colour plate used for colour vision tests, by
which a subject with normal vision can see ‘6’ that cannot be perceived by a
person without.
Figure 2.29. An example of the plate in the Ishihara colour vision test [Ishihara, 2004]. [29]
Elham Khodamoradi Page 62
2.8 Literature Review on chromatic texture
In this research, the initial observer training is conducted on CRT (cathode ray
tube) monitors to be in line with the existing studies [Gao, 2015]. In addition,
textured colours are studied to evaluate both subjects’ consistency and the
performance of the CIECAM02 model.
As a part of the human vision system that involves a light source, an object, and
a processing system (i.e. brain), colour vision is the ability of the human eye to
separate various frequencies of light waves (380~740nm) to allow us to
differentiate different colours through three colour cells, i.e. red, green, and blue,
positioned on the retina. Usually all colour models are established in a laboratory
environment, where, typically, the background is usually of a certain colour with a
uniform light source and test stimuli of single colours in a way that is different
from natural viewing situations.
2.8.1 Texture definition
Texture is a measure of the variation of the intensity of a surface, quantifying
properties such as smoothness, roughness and regularity. It is often used as a
region descriptor in image analysis and computer vision [Julesz,1962]. The three
principal approaches used to describe texture are statistical, structural and
spectral. Texture perception is important for colour perception and regarded as
one of the early steps towards identifying objects and understanding scene
through textural variations on the surface of an object. Largely attributed to
[Julesz, 1981] conjecture that observers were sensitive only to differences in the
first- and second-order statistics of constituent textures (though counter
Elham Khodamoradi Page 63
examples to this conjecture have been found at a later stage). Texture
segregation has been represented extensively using computational models with
a common form simulating cortical cells (simple or complex) including a set of
linear spatial filters, a point-wise nonlinearity, and further linear spatial filtering to
turn the textural difference into a difference in response strength.
These biologically inspired models accept grey-level images as an input, working
exclusively on the luminance domain. As a result, they concentrate on
differences in the figural, or form, features of the texture elements (texels), while
all the other attributes of the texels (luminance, colour, stereo disparity, etc.)
remain fixed.
However, recently study has shown that the interaction between colour,
luminance and orientation can be attributed to the texture process. Furthermore,
the typical experimental settings differ considerably from those in our natural
environment. It has been shown that colour sensitivity and appearance is
influenced by adaptation to the colour distribution of natural images [Webster,
1997].
2.8.2 Chromatic texture
Many natural objects are categorized not only by shape and colour, but also by
their particular chromatic textures. While the roles of colour and shape have
been well explored in object recognition, chromatic texture has not. In this
investigation, the roles of colour and texture and their interaction are studied, in
an object classification task using familiar objects.
Elham Khodamoradi Page 64
The use of joint colour-texture features has been a popular approach to colour
texture analysis. At present, it appears, there have not been many published
articles in this area which certainly shows colour and texture should be
processed jointly.
The issue has been touched on by a few scientists [Julesz, 1981], in terms of the
contribution of colour information to the classification of colour textures, but the
amount of data that has been considered is limited and the experiments take
place under limited conditions.
Therefore, in this research, Phase 1 is dedicated to study textured colours to
measure the significance of texture in contributing to the perception of colours,
specifically to estimation using the CIECAM02 model, whereas Phase 2 will
focus on the investigation of the modelling of the colour appearance on mobile
telephones.
Elham Khodamoradi Page 65
2.9 Literature Review on Smartphones
2.9.1 Introduction to Smartphone Global trends
Thanks to the advances of computer technology, our world at present is
becoming increasingly more mobile. Global sales of mobile devices have been
rising year by year, driven in particular by purchases in developing economies,
where the mobile device market is yet to reach a saturation point.
It was estimated that by the end of 2015 the global smartphone users would
reach 1.75 billion. One billion consumers had been recorded in 2012 [eMarket,
2015], which is demonstrated in Figure 2.27.
The total amount of mobile devices is already larger than PC desktop computers.
Global mobile device sales are not only focused on smartphones alone, but also
on feature phones which are classified as non-operating system devices. Since
these types are more affordable than current smartphones, these are still a
popular choice in countries such as India and China [eMarket, 2015].
Figure 2.30. ( Number of global users (2007-2015) [smartinsights.com, 2015][30]
Elham Khodamoradi Page 66
2.9.2 The rise of mobile internet worldwide
Mobile Internet users form a significant part of overall mobile subscriptions
worldwide. According to the estimation given by ITU, in 2013 there were 2.1
billion active mobile broadband subscriptions in the world. That means 29.5% of
the global population is online, and this number is sure to keep growing [Walk
digital, 2015].
According to the research by ‘We Are Social’, mobile overload has exceeded
100% in many regions of the world, including North America, Western Europe,
Central and South America, Central and Eastern Europe, and the Middle East
also. This means there are at least as many mobile subscriptions as citizens in
those regions, including users of all smartphones, feature phones, and regular
phones [WeAreSocial, 2015].
Figure 2.31: Comparison between mobile devices and fixed devices [IPcarrier, 2015]. [31]
Elham Khodamoradi Page 67
Figure 2.32: Device comparison (2013-2018) [CISCO, 2015]. [32]
Almost two-fifths of all mobile phone users, close to one-quarter of the worldwide
population, used a smartphone at least monthly in 2014. By the end of the
forecast period, smartphone penetration among mobile phone users globally will
near 50%.
Regarding the internet, more than 2.23 billion people worldwide, or 48.9% of
mobile phone users, went online via mobile at least monthly in 2014, and over
half of the mobile audience will use the mobile internet next year [Cisco, 2015].
As a result, it has been estimated that the total number of mobile phone internet
users rose 16.5% in 2014 and will maintain double-digit growth through 2016
[idc.com, 2015].
Elham Khodamoradi Page 68
2.9.3 Mobile and user expectations
User expectations are increasing in line with the introduction of more advanced
devices along with the huge exposure those consumers have with regard to
mobile websites and mobile apps, which contribute to the change of behaviour
patterns. Due to the change of users’ activities, digital mobility and connectivity
become ever more important every day as the smartphone continues to
transform what people demand. In a recent study by Salesforce, reported in
2014, 85% of respondents said that mobile devices were central to their
everyday life [Saleforce, 2015] as demonstrated in Figure 2.33.
Figure 2.33: Mobile usage comparison between UK and middle Europe and USA companies [Digital-stats, 2015]. [33]
Elham Khodamoradi Page 69
2.9.4 Smartphone UK trends
A smartphone is a mobile phone with more advanced computing ability and
connectivity than basic feature phones. Smartphones typically include the
features of a telephone, as well as the features from other popular user devices,
such as a personal assistant, a media player, a digital camera, and / or a GPS
navigation unit. Mobile telecommunications have been available in the UK since
the mid-1980s.
There are now 83.1 million mobile handsets and data connections in the UK. In
2000, just half of the UK adults said that they have a mobile phone. That figure
now stands at 93%. In the UK, there has been a steady but slowing growth of
smartphone usage; latest research from YouGov’s Technology and Telecoms
team confirms that almost half of mobile owners now use smartphones (47%)
[Portioresearch, 2015].
In the UK, it was believed that the number of smartphone users was expected to
reach 42 million people or 65% of the population [Mobilemastinfo, 2015]. Figure
2.34 projects this trend.
Elham Khodamoradi Page 70
Figure 2.34: The estimated growth of UK mobile smartphone users [Pharmaphorum, 2015]. [34]
2.9.5 Smartphone changing photography
The huge growth in popularity of smartphones in the past three years has led to
many consequences, some of which were easy to predict, and some of which
are not yet fully understood. This is true especially in the field of digital
photography. Smartphone users, it appears, take a lot of photographs.
The Apple iPhone 4 is currently the most popular 'camera' on Flickr.com [Flickr,
2015]. As a direct result, millions of dollars are being made by mobile developers
to develop photo apps to satisfy this new trend of demands.
2.9.6 Smartphone popularity
According to a new survey by market researcher Counter Point, the iPhone 5s
sold 7 million units in May, creating the bestselling smartphone in the world.
Secondly was Samsung's Galaxy S5, which sold 5 million units across 35
Elham Khodamoradi Page 71
countries. Table 2.3 lists the top 10 most popular smartphones on the market.
[Counter Point Research, 2014].
Table 2.3. Top 10 smartphone list as of May 2014 [Counter Point Research, 2014], [3]
Due to the adaptability and convenience of current mobile telephones, people
are constantly using a telephone to perform various tasks more than just making
a call, including viewing / taking pictures and shopping online. While a
smartphone can act as a mini-computer and can be utilized to search the internet
in the same way as using a desktop computer, it does not always offer the same
functionality as a computer due to its inherent limitations. For example, the
colour of a desktop computer can be adjusted to its desired colour temperature,
i.e. D65, whereas the RGB values on a smartphone normally cannot be modified
nor can the white balance be checked. As a result, performing online shopping
Rank Brand Model
1 Apple iPhone 5s
2 Samsung Galaxy S5
3 Samsung Galaxy S4
4 Samsung Note 3
5 Apple iPhone 5c
6 Apple iPhone 4S
7 Xiaomi MI3
8 Samsung Galaxy S4 mini
9 Xiaomi Hongmi Redrice
10 Samsung Galaxy Grand 2
Elham Khodamoradi Page 72
using a mobile phone can be tricky, especially when buying colour sensitive
items.
Therefore, this research takes an initiative to investigate the variations of colours
for a number of smartphones while making an effort to model their colour
appearance using CIECAM02. Phase 2 studies the Apple iPhone 5 as a starting
point but also considers the LG Nexus 4, Samsung, and Huawei models, and
makes a comparison with a CRT colour monitor, that has been studied
extensively in Phase 1 and that has been calibrated using the D65 standard.
In this study, the iPhone 5 (6), Samsung (1), Huawei (1), and LG Nexus 4(1) are
investigated due to their availability.
Elham Khodamoradi Page 73
3. Methodology
In this Chapter, experimental setup procedures, methodologies and the
apparatus applied for measuring colours are explained in detail, including the
setup of experiments and the magnitude estimation method. Table 3.1 gives an
overview of the experiments carried out in this study. In total, 19,510 data values
have been collected in this research.
Table 3.1 Overview of the psychophysical experiments carried out in the study. [4]
Phase Number
of
samples
Number
of
observers
Lighting
source
Device No of
Experiments
Scaling
attributes
No of
estimation
0
Training
30 7 D65 CRT 1 Lightness
Colourfulness
Hue
630
(=30x7x3)
1
Textured
Colours
30 7 D65 CRT
Textured
colours
5 Lightness
Colourfulness
Hue
3,150
(=30x7x3x5)
2
Mobile
phones
30 3-7 D65 IPhone5
(3)
LG
Nexus
Samsung
Huawei
6 Lightness
Colourfulness
Hue
3,780
(=30x7x6x3)
Total 7,560
Elham Khodamoradi Page 74
3.1 Apparatus
3.1.1 Viewing cabinet
Colours appear differently under different lighting sources. To avoid and reduce
the assessment error, the Verivide colour viewing cabinet is used to simulate
different light sources to obtain an objective assessment of colour and colour
difference under controlled viewing environments, as illustrated in Figure 3.1.
Figure 3.1: Colour viewing cabinet. [35]
3.1.2 Colour meter, CA-100
The phones and CRT were calibrated and measured before starting the
experiment. A Minolta Chroma Meter CS-100A, as shown in Figure 3.2, was
employed to measure each colour in terms of CIE tri-stimulus values, i.e. x, y, Y
and is in the position of the observers’ view.
Elham Khodamoradi Page 75
Figure 3.2: Colour meter, CA-100A. [36]
3.1.3 Monitoring calibration (Spyder)
Experiments are conducted on a 19” CRT monitor that was calibrated daily using
Pantone ColorVision OptiCAL Spyder software [Spyder]. Figure 3.3 illustrates
the calibration procedure using the Spyder software where the Gamma value is
set to 2.2.
Figure 3.3: Spyder sensor attached to the computer screen to perform the calibration. [37]
Elham Khodamoradi Page 76
The illuminant in these experiments is set to D65 to be consistent with the other
existing studies. Thirty test colours were selected to cover a wide range of
colours and were plotted in a CIE chromaticity xy diagram as given in Figure 3.4,
which is consistent with many existing studies [Gao, 2015].
Figure 3.4: Thirty test colours presented on a CIE xy Chromaticity diagram. The ‘*’ refers to reference white under D65 illuminant. [38]
Throughout all the experiments, the reference white, reference colourfulness and
surrounding colours remain the same. The test field in the centre in Figure 3.5
subtends a visual angle of 2o at a viewing distance of ~ 60 cm. Figure 3.5 shows
the difference of 2o and 4o vision defined by CIE when viewing them at a
distance of 45 cm.
The test field in the centre in Figure 3.6 subtends a visual angle of 2o at a
viewing distance of ~ 60 cm.
450 470
480
500
520
540
560
580
600
620
770
380 0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.2 0.4 0.6 0.8
y
x
The CIE x,y chromaticity diagram
Elham Khodamoradi Page 77
Figure 3.5: A viewing angle on 1931 CIE and 1964 CIE standard. [39]
3.2 Experimental setup
Figure 3.6 illustrates the experimental pattern applied in all the experiments.
Figure 3.6: The paradigm of the experimental pattern. [40]
Throughout each experiment, only the test sample in the centre changes from
sample 1 to sample 30, whereas the rest remains the same. The reference white
is assigned as having lightness of 100 whilst reference colourfulness has a
colourfulness of 40.
Elham Khodamoradi Page 78
Ten subjects with normal colour vision according to the Ishihara colour vision test
[Ishihara, 2004] were selected from 23 to conduct psychophysical experiments,
aged between 17 to 40 years old mixed with both males and females who have
different culture backgrounds and chosen from undergraduates, postgraduates
and post-doc researchers. Figure 2.14 shows an example of a colour plate used
for colour vision tests, by which a subject with normal vision can see a ‘6’ that
cannot be perceived by a person without.
3.3 Subject training
Since all the observers are new to the scaling experiment, a training experiment
was carried out first with a total of 23 subjects. Those who did not perform well
during the training experiment, showing big variations in their performance, were
not invited to take part in the subsequent formal experiments. The magnitude
estimation method as described in Section 2.4 is used to scale each colour in
terms of lightness, colourfulness, and hue. In other words, lightness represents
the degree of brightness a colour is showing. With the reference white being 100
and an imaginary black being zero, observers are asked to give an estimation
between 0 and 100 that is in proportion to the reference white. For example, if a
test colour appears half as bright as the reference white to a subject, 50% of
lightness should be assigned to the colour. On the other hand, colourfulness
shows the amount of hue appearing in a test colour, which is in proportion to the
reference colourfulness that is given colourfulness of 40, e.g. if a colour has a
colourfulness of 60, this colour should display 1.5 times as colourful as the
reference colourfulness colour.
Elham Khodamoradi Page 79
Neutral colours, including white, black, and grey colours have zero colourfulness.
For hue estimation, four opponent hue scales are employed, which are red
against green and yellow against blue. The hue of a test colour is then estimated
with reference to any two neighbouring colours. For example, a hue can appear
80% yellow and 20% red or 50% green and 50% blue. A colour cannot show the
contents of opponent hues, i.e. an answer of ‘30% green and 70% red’ is invalid
(red and green are opponent hues), whilst neutral colours have no content of hue
at all. This hue scale is later converted into a hue angle of 0 to 400 with 0 (400)
being red, 100 yellow, 200 green and 300 blue for data analysis.
Detailed descriptions are given below.
Lightness -- Lightness represents the degree of brightness a colour is showing.
Alternatively, lightness can be considered as an estimate of the lightness or
darkness of a colour. With the reference white being 100 and an imaginary black
being zero, observers are asked to give an estimation between 0 and 100 that is in
proportion to the reference white.
Colourfulness -- Colourfulness describes the purity and strength of a colour, e.g.
bright red or dull red. Colourfulness shows the amount of hue appearing in a test
colour, which is in proportion to the reference colourfulness that is given a
colourfulness of 40, e.g. if a colour has a colourfulness of 60, this colour should
display 1.5 times as colourful as the reference colourfulness colour.
Elham Khodamoradi Page 80
Hue -- Hue or shade is the colour name, e.g. red, yellow, blue. For hue
estimation, four opponent hue scales are employed, which are red against green
and yellow against blue. The hue of a test colour is then estimated with reference
to any two neighbouring colours. For example, a hue can appear 80% yellow and
20% red or 50% green and 50% blue. A colour cannot show the contents of
opponent hues.
To analyse these subjective data, arithmetic means for both hue and lightness
are applied as an averaged value between all the observers whereas a
geometric mean is employed for colourfulness as colourfulness is scaled using
an open-ended approach, which are defined as below.
Mean Value Definition: For a data set, the mean, or arithmetic mean, is the sum
of the values divided by the number of values, which is given in Eq. (3.1).
(3.1)
Geometric Mean Definition: The geometric mean is an average that is useful for
sets of positive numbers that are interpreted according to their product and not
their sum (as is the case with the arithmetic mean), and is given in Eq. (3.2).
(3.2)
Elham Khodamoradi Page 81
Standard Deviation Definition: A standard deviation refers to the measure of the
dispersion of a set of data from its mean. The more spread apart the data, the
higher the deviation. Standard deviation (s) is calculated as the square root of
variance and is formulated in Eq. (3.3).
(3.3)
CV Value Definition: In this study, the variations are expressed in CV values. The
coefficient of variation (CV) is defined as the ratio of the standard deviation 𝑠 to
the mean 𝑚, and is shown as Eq. (3.4).
𝐶𝑉 =𝑠
𝑚× 100% (3.4)
Elham Khodamoradi Page 82
4. Results on subject training and textured colours
4.1 Evaluation of Colour Monitor
The CRT Phillips Brilliance 201B colour monitor was calibrated and measured
every day during the period of experiment that lasted 12 months. A Minolta
Chroma Meter CS-100A was employed to measure each colour in terms of CIE
tri-stimulus values, i.e. x, y, Y from the position of an observers’ view. For each
experiment, although the test colours are different from the other experiments,
the reference white (RW), reference colourfulness (RC), and the background
should remain the same. The averaged colour difference of those measurements
in terms of CIELAB for reference white and reference colourfulness is 3 E*ab
units, which is similar to the literature [Luo ,1995], indicating the monitor is in a
consistent colour condition.
4.2 Evaluation of Subjects’ Data
Seven subjects were chosen for each experiment to have an age variation
between 20 to 40 years old and mixed gender, as given in Table 4.1. The
selection of the subjects aims to represent a wide age range group of subjects
while remaining comparable with studies in the literature [Luo, 2005].
Elham Khodamoradi Page 83
Table 4.1 Subject’s information. [5]
Subjects Age Gender
subject 1 31 Female
Subject 2 32 Female
Subject 3 28 Female
Subject 4 37 Male
Subject 5 19 Male
Subject 6 28 Male
Subject 7 36 Male
The first study therefore is trying to find out any difference between gender
groups. Figure 4.1 illustrates the comparison results for the estimation of
lightness, colourfulness, and hue between the averaged female and male
subjects. It can be seen from the graph that there are no significant differences
between gender groups since all the points are close to the middle straight line.
Elham Khodamoradi Page 84
Figure 4.1: The comparison results between female (x-axis) and male (y-axis) observers on the mean estimation of lightness (top), colorfulness (middle) and hue (bottom). [41]
Elham Khodamoradi Page 85
4.3 Experiments of textures
Thirty samples were displayed and estimated. In the first stage of the
experiment, the sets included 30 colour sample slides from a database. The
initial experiment (Experiment 1) is conducted on the 30 test colour samples
without any texture. The following six experiments add a number of grey textures
into those test colours as demonstrated in Figure 4.2. For example, in
experiment 2, a cross is added to the colours where the shade of the dot is the
same grey as the background.
Experiment 1 Experiment 2 Experiment 3
Experiment 4 Experiment 5 Experiment 6
Figure 4.2: An example of experimental patterns in the 6 colour and textured experiments where
Experiment 1 has no texture at all. [42]
In the second stage of the experiment, 5 sets of 30 colours with different textures
were tested, which aims to be similar to the texture database at VisTex
[VisTex,1995]. Figures 4.3 and 4.4 exemplify the experimental patterns in the
Elham Khodamoradi Page 86
real experimental settings whereas Figure 4.5 depicts the tristimulus values of 30
samples for all 6 experiments.
Figure 4.3: Experimental pattern for Experiment 1 (no texture) where only central test colour changes from sample 1 to sample 30. [43]
Figure 4.4: Textured colour samples (Experiment 2). [44]
Elham Khodamoradi Page 87
Figure 4.5: The tristimulus values (x, y) of thirty test colour samples for the six experiments conducted in Phase 1. The big red square in the centre represents the reference white under D65 that remains the same for all the experiments. [45]
As illustrated in Figure 4.5, the colour measurements do appear differently when
presented with different texture patterns but not as significant as the same colour
being presented on different telephones as will be seen in Figure 5.2 in Phase 2.
The reference white remains the same for all six experiments and for all 30
colour samples in each experiment. The six experiments use the texture patterns
that are demonstrated in Figure 4.2.
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.2 0.4 0.6 0.8
y
x
The CIE x,y chromaticity diagram
Exp-1
Exp-2
Exp-3
Exp-4
Exp-5
Exp-6
White
Elham Khodamoradi Page 88
4.4 User study
As described in Section 3, for calculation of the averaged visual estimates,
arithmetic means were employed for the attributes of lightness and hue, whilst
geometric means were used for colourfulness as colourfulness was scaled using
an open-ended approach. The coefficient of variation (CV) was applied to
measure the agreement between each individual’s estimation of the three
attributes against the actual values measured by the Chroma Meter, over the 30
test samples. Table 4.2 lists the mean CV values for the colour experiment and
Tables 4.3 lists the CV values for the different textured colour experiments.
Figure 4.6 shows figuratively the comparison between mean CV values for
Subject A (1), the rest of the figures are given in Appendix B. It shows the
variation of individual subject’s estimations in comparison with the mean
estimations for the attributes of lightness, colourfulness and hue.
Figure 4.6: Comparison of estimation between Subject 1 (A) and means. [46]
A total of 7 subjects were recruited and took part in Experiment 1. All 7 subjects
remained available to take part in Experiments 2 up to 6. Table 4.2 lists the CV
values for all 7 subjects in performing Experiment 1 which contains colour
samples without any texture, whereas the CV values for the rest of the
experiments, i.e. Experiments 2 to 6 are given in Appendix C.
0
20
40
60
80
100
0 20 40 60 80 100Mean
Colorfulness
Elham Khodamoradi Page 89
Table 4.2. The CV values of all seven subjects who performed Experiment 1 (Colour Sample). [6]
Subject
Lightness Colourfulness Hue
CV r a b CV r a b CV r a B
A 20 0.88 1.05 -2.7 32 0.62 1.6 1.15 9 0.95 1.04 -17.7
B 17 0.87 0.76 11.8 26 0.44 0.54 18.23 11 0.94 1.06 -15.6
C 36 0.89 1.18 -21.9 26 0.61 0.789 4.05 13 0.92 1.08 15.67
D 41 0.89 1.14 -11.2 20 0.81 1.14 -6.72 9 0.96 0.95 12.92
E 29 0.62 0.58 29.78 20 0.64 1.06 7.94 14 0.93 1.02 -11.7
F 25 0.91 1.26 -3.21 16 0.64 0.76 6.83 16 0.95 1.07 -34.3
G 23 0.9 1.18 -10.3 63 0.60 0.779 25.66 9 0.95 1.06 2.3
Mean 27 0.82 1 0.003 29 0.62 0.902 12.72 11 0.90 1.466 0.01
As shown in Table 4.2, each estimate by the subject is measured using three
values, 𝑎, 𝑏 and 𝑟 (correlation coefficient). A regression equation allows us to
find out the relationship between two (or more) variables to be expressed
algebraically. It indicates the nature of the relationship between two (or more)
variables.
A linear regression equation is usually written as Eq. (4.1):
𝑌 = 𝑎𝑋 + 𝑏 (4.1)
Elham Khodamoradi Page 90
where (𝑎) is the slope and (𝑏) is the intercept and 𝑟 is the independent variable to
measure the linear relationship between datasets X and Y, which lies within -1 to
1.
For example, for Subject 1 (A in Table 4.2), the linear equation is written and
calculated as shown below.
Figure 4.7: Plot between mean lightness (x) with lightness estimation by Subject 1 (y). [47]
Therefore, in this case, 𝑎 = 1.05; 𝑏 = −2.74; and 𝑟 = 0.88. In the ideal perfect
situation, 𝑎 should be 1 and 𝑏 should 0 and 𝑟 should be 1.
In Table 4.2, according to the CV values, all subjects perform well on hue
estimations with the average of 11% errors and worst on the estimation of
colourfulness with the average error of 29%. The average error on the estimation
of lightness is 27%. In comparison with the other studies [Luo 1993a, Luo 1995],
which have errors of 17%, 25% and 10% for the estimation of lightness,
colourfulness and hue respectively, the errors are very similar for colourfulness
and hue. The estimation of lightness is much larger in this study with 27% error.
This is mainly because most of the subjects were new to the magnitude
estimation technique whereas in Luo’s studies, those subjects are very
Elham Khodamoradi Page 91
experienced observers. For example, for Subject D and Subject G, their
estimation of lightness and colourfulness respectively constitute the largest
errors, which contribute to the final mean values.
On the other hand, although the CV value for colourfulness is 29%, which is
similar to Luo’s study, the correlation coefficient (𝑟) is only 0.62 whereas in Luo’s
study, it is greater than 0.8, implying that for this study the estimation has larger
scattering around the central line.
In this study, the subjects become more skilled during the subsequent
experiments. As the subjects become more experienced, they perform better
with decreasing CV values. For example, in Table 4.3, the CV values are
improved significantly for lightness estimation, i.e. from 29% down to 21%, which
is consistent with those in Luo’s studies. In particular, the r value for
colourfulness in Table 4.3 has increased to 0.77 from 0.62, implying that
observers have much less scattering estimations. Table 4.4 summaries CV
values for all subjects while conducting the six experiments whilst Appendix A
lists all the CV values for each individual experiment.
Elham Khodamoradi Page 92
Table 4.3. CV values for Experiment 2 (textured Colour Sample). [7]
Subject
Lightness Colourfulness Hue
CV r a b CV r a b C
V r A b
A 24 0.94 1.14 -
15.4 29 0.72 1.00
-
1.04 11 0.74 0.91
-
4.74
B 17 0.95 1.13 -
9.52 31 0.79 1.1
-
7.65 11 0.81 1.05
-
24.2
C 14 0.93 0.9 3.43 20 0.88 1.06 -
4.77 14 0.89 1.19
-
16.3
D 24 0.84 1.05 -
3.44 23 0.79 0.92
-
0.96 13 0.7 0.85 37.2
E 27 0.74 0.73 12.4 27 0.77 0.94 5.66 14 0.91 1.23 -
51.2
F 23 0.88 1.08 3.38 43 0.74 0.93 16.5 15 0.74 1.94 33.9
G 22 0.9 0.94 9.09 43 0.7 0.68 24.7 18 0.74 0.87 25.4
Mean 21 0.88 30 0.77 13
Elham Khodamoradi Page 93
From Table 4.3, it can be seen that Hue is the easiest attribute to estimate with
the least CV values, ranging from 11% to 15%, whereas colourfulness is the
most difficult one with CV values ranging from 26% to 30%. The CV values for
lightness estimation are between 21% to 27%. Overall, all the subjects seem to
perform to a similar standard, which is as expected.
Elham Khodamoradi Page 94
Table 4.4. Summary of the CV values for Subjects A to G while performing
Experiments 1 to 6. [8]
Subject A B C D E F G Mean
Exp-1
L
24 17 36 41 29 25 23 27
C 32 26 26 20 20 16 63 29
H 9 11 13 9 14 16 9 11
Exp-2
L
24 17 14 24 27 23 22 21
C 29 31 20 23 27 43 43 30
H 11 11 14 13 14 15 18 13
Exp-3
L
38 22 27 53 32 12 29 30
C 49 30 28 40 34 21 29 33
H 11 11 7 19 11 8 16 11
Exp-4
L
20 17 36 41 29 25 23 27
C 32 26 26 20 20 16 28 24
H 9 11 13 9 14 16 9 11
Exp-5
L
26 21 20 27 37 18 19 24
C 26 38 24 22 30 32 29 28
H 18 24 12 12 16 12 14 15
Exp-6
L
17 16 17 17 12 29 22 18
C 21 20 35 24 30 31 26 26
H 15 15 11 10 13 14 19 13
Elham Khodamoradi Page 95
Table 4.5 compares the CV values across all six experiments. Again, a similar
tendency occurs, i.e. colourfulness is the most difficult attribute to estimate.
Specifically, as the experiments progress, the CV values are getting smaller for
both lightness (from 27% to 22%) and colourfulness (from 29% to 18%),
suggesting that subjects are getting more experienced and producing more
consistent estimations.
Table 4.5. Summary of the mean CV (%) values of subjects’ estimations for the 6
experiments. [9]
Name Lightness Colourfulness Hue
Experiment 1 27 29 11
Experiment 2 21 30 13
Experiment 3 30 33 11
Experiment 4 27 24 11
Experiment 5 24 28 15
Experiment 6 22 18 13
Mean 25 27 12
Elham Khodamoradi Page 96
4.5 The effect of texture on the appearance of
colours
Experiments 2 to 6 are performed on textured colour samples as illustrated in
Figure 4.2. A graphical comparison of the variability between the sets of samples
in terms of the three attributes is shown in Figure 4.8 where comparison between
Experiment 1 (x-axis without texture) and the other 5 experiments are given.
0
20
40
60
80
100
0 20 40 60 80 100
Me
an L
igh
tne
ss
Mean from Experiment 1
Lightness
lightness-exp-2
lightness-exp-3
lightness-exp-4
lightness-exp-5
lightness-exp-6
Elham Khodamoradi Page 97
Figure 4.8: Comparison results between mean estimations of lightness (top), colourfulness (middle) and hue (bottom) for Experiment 1 (x-axis) and Experiments 2-6 (y-axis). [48]
As can be seen from Figure 4.8, the largest discrepancies occur on colourfulness
estimations, in particular for Experiments 4 () and 6 (), which contain patterns
of random lines and random dots as depicted in Figure 4.2, which could be
0
20
40
60
80
100
0 20 40 60 80 100
Me
an C
olo
rfu
lne
ss
Mean from Experiment 1
Colorfulness
colorfulness-exp2
colorfulness-exp-3
colorfulness-exp-4
colorfulness-exp-5
colorfulness-exp-6
0
100
200
300
400
500
0 100 200 300 400 500
Me
an H
ue
Mean from Experiment 1
Hue
hue-exp-2
hue-exp-3
hue-exp-4
hue-exp-5
hue-exp-6
Elham Khodamoradi Page 98
explained on the basis that random patterns make colourfulness more difficult to
estimate. Although Experiment 2 also contains patterns scattered around the
colour sample, the texture pattern itself, i.e. X, is symmetrical. Otherwise, all the
points in Figure 4.8 are slightly above the central line for the colourfulness
estimation, suggesting that texture has some effect on colourfulness estimations
and makes colours appear to be more colourful.
For lightness, the textured colours appear to be darker than the colours without
textures, which indicate the texture decreases the lightness contrast and makes
colours appear darker.
On the other hand, little effect is found for hue estimation, which is expected as
the texture patterns are neutral colour, the same grey as the background.
Therefore it is concluded in this study that texture that shares the same grey
level with that in the background has little effect on hue estimation, makes a
colour appear more colourful and darker, while under the D65 viewing condition
on CRT monitors.
A comparison between texture patterns was also carried out, which is
demonstrated in Figures 4.9 where estimation results from Experiment 6 (with
dot texture, x-axis) are compared with the results from Experiments 5 (with cross
texture) and 4 (with line texture) respectively.
Elham Khodamoradi Page 99
Figure 4.9: Comparison results between Experiment 5 (dot pattern, x-axis)) with Experiment 6 (cross pattern) and Experiment 4 (line pattern). [49]
The figure shows the estimation results are very similar, implying that texture
patterns have little effect on the colour appearance in this study. Similar results
0
20
40
60
80
100
0 20 40 60 80 100
Me
an -
Cro
ss
Mean-Dot
Lightness
0
20
40
60
80
100
0 20 40 60 80 100
Me
an -
Lin
e
Mean-Dot
Lightness
0
20
40
60
80
100
0 20 40 60 80 100
Me
an -
Cro
ss
Mean-Dot
Colorfulness
0
20
40
60
80
100
0 20 40 60 80 100
Me
an -
Lin
e
Mean-Dot
Colorfulness
0
100
200
300
400
500
0 100 200 300 400 500
Me
an-C
ross
Mean-Dot
Hue
0
100
200
300
400
500
0 100 200 300 400 500
Me
an-L
ine
Mean-Dot
Hue
Elham Khodamoradi Page 100
can also be found when comparing between patterns of crosses (Experiment 5)
and lines (Experiment 4) as depicted in Figure 4.10.
Figure 4.10: Comparison results between Experiment 4 (line pattern, x-axis) with Experiment 6 (cross pattern). [50]
From Section 4.5, it can be seen that the added texture appears to have some
effects on estimated colour appearance. Therefore this analysis focuses on the
study of the performance of CIECAM02 on the prediction of these colour
appearances. The graphs in Figures 4.11 illustrate the comparison results
between observers’ estimation and the predictions by CIECAM02. That is the
data estimated by subjects (x-axis) from each experiment is plotted against the
0
20
40
60
80
100
0 20 40 60 80 100
Me
an C
ross
Mean-Line
Lightness
0
20
40
60
80
100
0 20 40 60 80 100
Me
an C
ross
Mean-Line
Colorfulness
0
100
200
300
400
500
0 100 200 300 400 500
Me
an-C
ross
Mean-Line
Hue
Elham Khodamoradi Page 101
predications by CIECAM02 for each of three colour attributes of lightness,
colourfulness, and hue. The input data of XYZ of CIECAM02 are measured in
advance for each colour sample under each experimental setting as presented in
Figure 4.5.
With regard to hue, the predictions from CIECAM02 fit well with subjects’
estimations. Because of the presence of grey texture patterns presented in each
test colour sample, the estimations of lightness by subjects appear to be higher,
which could be due to the fact that the presence of texture makes colours appear
darker. On the other hand, the estimated colourfulness appears to be less
colourful than that of predictions by CIECAM02 due to the fact that textures
make colour more colourful as observed in Section 4.5. Note: Experiment 1 does
not have any texture pattern; therefore the top left graph appears to be a good fit.
With regard to the modelling of textured colours using CIECAM02, a linear
modification has been carried out on Eqs. (4.2) and (4.3) for both lightness and
colourfulness respectively as it gives the biggest correlation coefficient values
compared to other non-linear approaches.
𝐿𝑡𝑒𝑥𝑡𝑢𝑟𝑒 = 0.82 𝐿𝐶𝐼𝐸𝐶𝐴𝑀02 (4.2)
𝐶𝑡𝑒𝑥𝑡𝑢𝑟𝑒 = 1.40 𝐶𝐶𝐼𝐸𝐶𝐴𝑀02 (4.3)
In this way, the average correlation coefficient (r) values are 0.86 and 0.61 for
lightness and colourfulness respectively. Obviously textured colours have more
complex characteristics than linear representations. More study will be
conducted in the future to investigate more textured patterns.
Elham Khodamoradi Page 102
The same experiment on colour samples was also conducted on mobile phones
and as the result of data comparison and fitting CIECAM02 model to this phase
of the study, the conclusion and changes in CIECAM02 model is as follows.
Since the colour appearance model CIECAM02 was developed for the medium
of a CRT, which feature refection and transparency, it is not well equipped to
predict smartphones. After setting the environmental parameters to ‘dim’
conditions where F = 0.9, c = 0.59, and Nc = 0.90 to compensate for lightness
differences between a CRT and iPhone, the comparison results are given in
Figure 5.6 for the three iPhones, where colourfulness was adjusted according to
Eq (4.4) below.
Colourfulness_smartphone = 1.8 * Colourfulness_CIECAM02 (4.4)
When comparing with the other smartphones, in terms of hue, all three types of
phones tend to be more reddish for purplish colours than those displayed on an
iPhone, whereas the rest maintains near the same. With regard to lightness, for
lighter colours, all three phones unanimously appear darker than on iPhones with
the LG-Nexus being the darkest with 25% darker, whereas 18% and 16% darker
are evidenced for HuaWei and SamSung respectively. However, the opposite
phenomenon occurs when it comes to the representation of colourfulness.
The images on all three phones, the LG-Nexus4, SamSung, and HuaWei,
appear more colourful than those displayed on an iPhone. For example, the
colours on a LG-Nexus4 phone appear systemically 10% more colourful than
those depicted on an iPhone. In addition, for both the HuaWei and SamSung
Elham Khodamoradi Page 103
phones, they appear again to be 17% and 22% more colourful, especially for
colourful samples, the tendency presenting across all three smartphones.
Elham Khodamoradi Page 104
4.6 Data prediction using CIECAM02
From Section 4.5, it can be seen that the added texture appears to have some
effects on estimated colour appearance. Therefore this analysis focuses on the
study of the performance of CIECAM02 on the prediction of these colour
appearances. The graphs in Figures 4.11 illustrate the comparison of results
between observers’ estimation and the predictions by CIECAM02. That is the
data estimated by subjects (x-axis) from each experiment is plotted against the
predications by CIECAM02 for each of three colour attributes of lightness,
colourfulness, and hue. The input data of XYZ of CIECAM02 are measured in
advance for each colour sample under each experimental setting as presented in
Figure 4.5.
Understandably, CIECAM02 is not designed for textured colour samples.
Therefore its predictions may vary to a large extent. Interestingly, with regard to
hue, the predictions from CIECAM02 fit well with subjects’ estimations. Because
of the presence of grey texture patterns presented in each test colour sample,
the estimations of lightness by subjects appear to be higher, which could be due
to the fact that the presence of texture makes colours appear darker. On the
other hand, the estimated colourfulness appears to be less colourful than that of
predictions by CIECAM02 due to the fact that textures make colour more
colourful as observed in Section 4.5. Note: Experiment 1 does not have any
texture pattern, therefore the top left graph appears to be a good fit.
Elham Khodamoradi Page 105
Figure 4.11: The prediction of lightness by CIECAM02 (y-axis) are compared with subjects’ estimations of lightness for Experiments 1 to 6 (x-axis). [51]
0
20
40
60
80
100
0 20 40 60 80 100
CIE
-Lig
htn
ess
Exp-4-2
CIE Lightness vs Exp-4
Elham Khodamoradi Page 106
Figure 4.12: The prediction of Colourfulness by CIECAM02 (y-axis) are compared with subjects’ estimations of colourfulness for Experiments 1 to 6 (x-axis). [52]
Elham Khodamoradi Page 107
Figure 4.13: The prediction of Hue by CIECAM02 (y-axis) are compared with subjects’ estimations of Hue for Experiments 1 to 6 (x-axis). [53]
0
100
200
300
400
500
0 100 200 300 400 500
CIE
-Hu
e
Exp-1
CIE Hue vs Exp-1
0
100
200
300
400
500
0 100 200 300 400 500
CIE
-Hu
e
Exp-2
CIE Hue vs Exp-2
0
100
200
300
400
500
0 100 200 300 400 500
CIE
-Hu
e
Exp-3
CIE Hue vs Exp-3
0
100
200
300
400
500
0 100 200 300 400 500
CIE
-Hu
e
Exp-4-2
CIE Hue vs Exp-4-2
0
100
200
300
400
500
0 100 200 300 400 500
CIE
-Hu
e
Exp-5
CIE Hue vs Exp-5
0
100
200
300
400
500
0 100 200 300 400 500
CIE
-Hu
e
Exp-6
CIE Hue vs Exp-6
Elham Khodamoradi Page 108
Because the calculation of Hue is cycling between 0-400, i.e. 420 is equivalent to
20, the same data in Figure 4.11 appears to be scattered widely in comparison
with Figures 4.9 and 4.10, whereas in fact the scattering is smaller.
In summary, the data collected from experiments 1 to 6 are based on 30 test
colour samples estimated by 7 subjects after being trained and under a
controlled environment by measuring and visual estimations of lightness,
colourfulness and hue. The comparison takes place between the mean values of
each attribute and the corresponding ones predicted by the CIE colour
appearance model, i.e. the CIECAM02 model.
In Experiment 1, all lightness, colourfulness and hue values estimated by the
observers appear to agree well with CIECAM02 predictions, indicating the
experiments reported in this thesis are in line with the published work.
For experiments 2 to 6, a variety of simple texture patterns have been added to
the test colour samples. The comparison results between the predictions by
CIECAM02 and the observers’ estimations show that they agree very well for the
attributes of hue. For lightness, the predictions appear to be darker, whilst for
colourfulness, the predictions appear to be more colourful, which is in line with
the results Section 4.5.
4.7 Summary of Phase 1
The added texture does make changes to the appearance of a colour. Although
the hue content does not change, a colour does appear darker and slightly more
Elham Khodamoradi Page 109
colourful when it has a texture in it than without. The pattern of the textures has
no effect on the appearance of a colour.
The predictions by CIECAM02 follows the same trend, the textured colour
samples appearing darker and more colourful.
Phase 2 of this study focuses on conducting similar experiments on mobile
telephones whereas Phase 1 was conducted on a CRT monitor. Again,
CIECAM02 will be investigated for the fitness to subjects’ estimations.
Elham Khodamoradi Page 110
5. Results on smartphones
5.1 Summary of the experimental setting
This study introduced refined versions of CIECAM02 for mobile displays. The
images were processed based on CIECAM02 phenomena (lightness,
colourfulness and hue). The experiment was conducted by comparing the
original images viewed in a viewing cabinet with the predicted image viewed on
mobile displays. The refined CIECAM02 with colourfulness* correction performed
much better than original CIECAM02 versions. Many colour appearance studies
were carried out using household TV or PC displays viewed under controlled
viewing conditions. However, the colour appearance of mobile displays is
affected by a many of viewing conditions. First of all, the display size is much
smaller than the other displays such as TV. Secondly, the portability and the size
of the device allows the display to be viewed under surrounding conditions
varying from day light to bright sunlight, and indoor and outdoor conditions.
This phase of the study, describes the modelling of a few well known smart
phones by using coloured image samples based on individual image statistical
measurements for mobile phone displays (LCD). Thirty natural images were
measured and analysed in terms of lightness, colourfulness and hue attributes.
Each of the images was displayed on a mobile LCD and assessed by a panel of
10 observers. The visual results were used to evaluate the CIE colour
appearance model, CIECAM02. The model was then modified specifically for
mobile display viewing conditions.
Elham Khodamoradi Page 111
Colour has been recognised as one of the top considerations in recognising an
image. Human colour recognition is psychological, and is therefore subjective,
hence it can be classified according to human visual perception. Objective
evaluation involves physical measurement of images but generally fails to
consider human visual characteristics. Therefore, psychophysical experimental
results are required for developing metrics.
A number of these metrics have been suggested and widely used such as the
CIELAB colour difference equation and CIECAM02 was developed in 1996 as an
image difference metric factor for colour image properties.
In phase 1, affective attributes in modelling of colour image followed by colour
textured image were investigated. These included lightness, colourfulness and
hue measurement of the samples by observers. The experiment was designed to
develop two types of colour and textured samples.
In phase 2, only colour attributes were considered and the statistical values of
images were used to develop a colour model.
Although an image begins by being represented using RGB colour space when it
is in a digital form, it is usually converted into hue, lightness and colourfulness
colour space to circumvent the dependency nature of RGB space on hardware
devices, i.e. a colour in one device usually does not appear nor measure the
same as one in another device even with the same RGB values in both devices.
This is because the range of R, G, or B values are manually set to be the same
(such as [0, 255] for an 8-bit computer) for all monitors regardless of their
physical measurements. On the other hand, hue, colourfulness and lightness,
space agrees more with human vision theories. To further improve the fitness
Elham Khodamoradi Page 112
between users’ perception and retrieved results, colour appearance based
retrieval has been proposed in [Qui, 2002], attention is paid to the development
of distance formulae, whereas in [Othman, 2008], patterned colour is the focus.
Usually, colour effect can be subjectively specified by means of visual perception
[Judd, 1965]. By the definition given by the CIE (Commission Internationale de
l'éclairage) [CIE: Publication No 17,1970], the hue, colourfulness and lightness,
abstracted from complete visual experiences, are employed to represent
dimensions along which colour may vary independently. For example, Hue is
defined as the attribute of a visual sensation according to which an area appears
to be similar to one, or to proportions of two, of the perceived colours, red,
yellow, green, and blue, whereas Colourfulness constitutes the attribute of a
visual sensation according to which an area appears to exhibit more or less of its
hue. Brightness implies the attribute of a visual sensation according to which an
area appears to exhibit more or less light, and Lightness refers to the brightness
of an area judged relative to the brightness of a similarly illuminated area that
appears to be white or highly transmitting [CIE: Publication No. 17, 1970].
In this Chapter, experimental setup procedures, methodologies and the
apparatus applied for measuring colours are explained in detail, including the
setup of experiments and the magnitude estimation method.
Thirty test colours are randomly selected from the Munsell colour book [Munsell]
while making an effort to cover as much of the CIE 1931 colour space as
possible. Psychophysical experiments are then carried out on both a 19” CRT
colour monitor with its illuminant calibrated to D65, and mobile telephones. As
illustrated in Figure 5.1, each test colour is placed at the centre against a grey
Elham Khodamoradi Page 113
background (with 20% of the luminance of reference white) and surrounded by
the reference white, reference colourfulness and surrounding colours. The test
field in the centre subtends a visual angle of 2o at a viewing distance of ~60cm.
Ten subjects with normal colour vision are selected to conduct the experiments
using the technique of magnitude estimation which they have been trained in
advance to apply skilfully. Specifically, for each test colour, each subject is asked
to estimate its appearance in terms of lightness, colourfulness and hue contents
verbally that are then recorded by an operator sitting nearby.
Figure 5.1: The paradigm of the experimental pattern. [54]
The above paradigm has been used to display on the mobile telephones and the
tests were carried out by using the coloured sample on each mobile telephone
using the same experimental procedure as phase 1. Different colours were
selected in different samples to give a practical coverage of the CIELAB space.
Neutral grey background colours were used in each phase to clearly show the
Elham Khodamoradi Page 114
colours on the monitor with the right contrast. The sample was placed inside a
viewing booth and the observers with normal colour vision, according to the
Ishihara test, participated in all experimental phases. All observers were familiar
with the magnitude estimation method. Each was asked to estimate test colours
in terms of lightness, colourfulness and hue closely following the method used by
Luo et al [Luo, 1995].
Lightness was scaled against the reference white having a lightness of 100 and
an imaginary black, 0. An anchor patch that was assigned a colourfulness of 40
and brightness of 100 was shown in a viewing cabinet. Each observer had to
estimate the colourfulness of the reference patch in the beginning of each phase.
The brightness was judged according to the anchor patch having a value of 100.
The hue was judged by reporting the percentage of the two colours from the four
psychological hues (red, yellow, green, and blue). The measured data are
presented in a CIE xy chromaticity diagram as illustrated in Figure 5.2.
Figure 5.2: The x, y data of the thirty sample colours plotted on the CIE xy chromaticity diagram for the four telephones studied in this research. The big * refers to the reference white. [55]
Elham Khodamoradi Page 115
5.1.1 Smartphones Colour Gumet
The measured data are presented in a CIE xy chromaticity diagram for each
individual smart phone as illustrated in Figure 5.3.
Figure 5.3: The x, y data of the thirty sample colours plotted on the CIE xy chromaticity diagram
for the three iPhones studied in this research. The big * refers to the reference white. [56]
Figure 5.4: The x, y data of the thirty sample colours plotted on the CIE xy chromaticity diagram
for the LG-Nexuse4 studied in this research. The big * refers to the reference white. [57]
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.2 0.4 0.6 0.8
y
x
The CIE x,y chromaticity diagram
iPhone-1
iPhone-2
iPhone-3
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.2 0.4 0.6 0.8
y
x
The CIE x,y chromaticity diagram
LG-Nexus4
Elham Khodamoradi Page 116
Figure 5.5: The x, y data of the thirty sample colours plotted on the CIE xy chromaticity diagram
for the HuaWei studied in this research. The big * refers to the reference white. [59]
Figure 5.6: The x, y data of the thirty sample colours plotted on the CIE xy chromaticity diagram for the Samsung studied in this research. The big * refers to the reference white. [60]
In addition, comparison between subjects’ estimation is also conducted,
particularly with regard to hue, the estimated hue values of the colours
presenting on a mobile telephone appear to be similar between all telephones
and between phones and the CRT monitor. Similarly, the colourfulness and
0.00.20.40.60.81.0
0.0 0.2 0.4 0.6 0.8
y
x
The CIE x,y chromaticity diagram
HuaWei
0.0
0.2
0.4
0.6
0.8
1.0
0.0 0.2 0.4 0.6 0.8
y
x
The CIE x,y chromaticity diagram
SamSung
Elham Khodamoradi Page 117
lightness vary significantly on subjects’ estimations. For the iPhone 5, double
values of lightness and double colourfulness have been witnessed. As a result,
the CIECAM02 model has been modified to convert an image taken by a mobile
telephone to its colour corrected using both forward and backward models. In
doing so, an image is firstly converted into XYZ representation using the
correlation between the RGB space and XYZ from a mobile telephone that has
been measured in advance. Then using the forward model of CIECAM02, the
predictions of lightness, colourfulness and hue are obtained.
After the alteration of these values, i.e. halving (or doubling if showing on a
computer monitor) the lightness and colourfulness values of the image, the final
XYZ values of the image can be calculated using the reverse model of
CIECAM02. Subsequently, the RGB values and the final image will be displayed
with truthful colours.
5.1.2 Observer Variations
Observer variations in terms of observer accuracy were examined. The accuracy
was compared between each individual observer and mean visual results. The
measure of Coefficient of Variation (CV) was used to indicate the disagreement
between two sets of data. The more the points are scattered about the line, the
poorer the agreement. For a perfect agreement, CV should be zero and the
larger the value, the poorer the agreement.
The Figure 5.7 below shows the above sample on the mobile telephone screen
which had been used for all the subjects during the test.
Elham Khodamoradi Page 118
Figure 5.7: Sample test on mobile phone screen. [61]
5.1.3 Setup on a Smartphone
Test samples were displayed on a mobile telephone. The colour gamut is similar
to RGB as shown in the CIE 1931 xy chromaticity diagram. A Minolta CS-1000
telespectro radiometer was used for measurement.
The CIECAM02 model is the colour appearance model recommended by CIE.
The model’s performance is evaluated using the present experimental data in
terms of CV calculated between the model’s predicted and visual results.
5.1.4 Apparatus
Colours appear differently under different lighting sources. To avoid and reduce
the assessment error, a colour viewing cabinet is used to simulate different light
sources to obtain an objective assessment of colour and colour difference under
Elham Khodamoradi Page 119
controlled viewing environments, as illustrated in below Figure 5.8 (a) and (b).
(a)
(b)
Figure 5.8 (a,b): Mobile telephone under viewing cabinet position. [ 62]
Elham Khodamoradi Page 120
Figure 5.9 below shows the mobile telephone experimental procedure which was
carried out by 10 subjects under viewing cabinet conditions and demonstrate the
subject position and the environmental situation during the experiment.
Figure 5.9: Subject’s viewing position. [63]
The telephones and CRT were calibrated and measured before starting the
experiment. A Minolta Chroma Meter CS-100A, as shown in Figure 5.10, was
employed to measure each colour in terms of CIE tri-stimulus values, i.e. x, y, Y
and was in the position of the observers’ view.
Figure 5.10: Colour meter, CA-100A. [64]
Elham Khodamoradi Page 121
In addition, each colour on each phone has been measured using a colour meter
CS-100A, simulating the subjects’ viewing position. In particular, the reference
white, reference colourfulness, and background were measured at least 3 times,
e.g. at the beginning, the middle and the end of the colour sample sequence, to
check the repeatability of the telephone. In total, 6 iPhone 5s, 1 Huawei, 1
Samsung, 1 LG Nexus4, are measured and estimated. The same work is also
performed on the CRT monitor, Philips Brilliance 201B.
5.2 Comparison of Experimental Data
In Phase 2, the experimental setting is the same as Experiment 1 in Phase 1 in
terms of measurement using the CS-100A colour meter on a series of colour
samples displayed on each telephone. Encouragingly, the variations among the
same kind of telephone are not significant with less than 3 E CIELAB units
when measuring reference white and background of a test pattern. Figure 5.3
demonstrates the comparison results between 2 iPhones using CIELUV L*C* as
formulated by Eq. (2.6) since CIELAB is mainly used for the calculation of colour
differences.
Elham Khodamoradi Page 122
Figure 5.11: The comparison results between 2 iPhones using CIELUV L* and C*. [65]
It can be seen from Figure 5.11 that all the data points are located around the
central line for both L* and C*, suggesting that the 2 iPhones have similar colour
attributes.
Figure 5.12 compares the measurement of each telephone (i.e. xyY values using
Colour Meter CA-100) represented using CIELUV L*C* with the measurement of
the CRT applied in Phase 1.
Figure 5.12: The comparison of CIELUV L* and C* between the SamSung phone and CRT. [66]
Elham Khodamoradi Page 123
Figure 5.12 shows that the L* of the SamSung is larger than that of the CRT
monitor. For C*, the less colourful colours have bigger C* and more colourful
colours have less C* on the SamSung telephone. Similar results can also been
found in Figure 5.13 where 2 mobile iPhones are compared with the CRT
monitor.
Figure 5.13: The comparison of CIELUV L* and C* between 2 iPhones and CRT. [67]
Figure 5.13 reiterates the effect that is found in Figure 5.4, i.e. the L* is larger for
both iPhones than the L* for the CRT. Also for most colours, the C* is larger for
Elham Khodamoradi Page 124
iPhones than for the CRT, suggesting colours on mobile telephones might
appear lighter and more colourful than on CRTs.
In addition, Figure 5.14 exemplifies the estimation results by subjects between
the CRT and an iPhone.
Figure 5.14: Comparison of the estimation results by subjects between CRT (x-axis) and iPhones
1 studied in this research. [68]
The above figure illustrates the colours on the iPhone appear lighter and slightly
more colorful than on CRT monitors.
For lightness, the estimation on a mobile iPhone tends to be 16% more than that
on CRT monitors, whereas an 11% increase of colourfulness for mobile
telephones is evidenced. In Figure 5.15, a hue-colourfulness plot is presented,
where small circles (o) represent the colours from the CRT and big squares (□)
from an iPhone 5.
Elham Khodamoradi Page 125
Figure 5.15: Subjects’ estimation for the same group of colours on both CRT and iPhone5. A small circle (o) refers to the estimation for the CRT whereas big solid square (■) on iPhone. [69]
With regard to hue estimations, although the correlation for Hue remains the
highest (i.e. r = 0.96) between the iPhone and CRT, i.e. greenish colours appear
to be greener and blue colours bluer as shown in Figure 5.15. All colours tended
to be more colourful while depicted on the telephone. Figure 5.16 demonstrates
visually the differences between the iPhone (top) and the CRT (bottom) when a
colour checker image is displayed.
Elham Khodamoradi Page 126
Figure 5.16: Colour checker appears on both CRT (bottom) and iPhone 5 (top). [70]
As pointed by arrows, the blueish colours appear differently when displayed on
different media.
Elham Khodamoradi Page 127
6. Modelling of iPhone Appearance using CIECAM02
Since the colour appearance model CIECAM02 was developed for the medium
of a CRT, which features refection and transparency, it is not well equipped to
predict smartphones. After setting the environmental parameters to ‘dim’
conditions where F = 0.9, c = 0.59, and Nc = 0.90 to compensate for lightness
differences between a CRT and iPhone, the comparison results are given in
Figure 6.1 for the three iPhones, where colourfulness was adjusted according to
Eq. (6.1).
Colourfulness_smartphone = 1.8 * Colourfulness_CIECAM02 (6.1)
Phone The Experiments
1
r = 0.81 r = 0.72 r = 0.95
2
r = 0.85 r = 0.73 r = 0.97
Elham Khodamoradi Page 128
3
r = 0.80 r = 0.59 r = 0.88
Figure 6.1: Comparison of CIECAM02 predictions with subject’s estimations. [68]
After correction using Eq. (6.1), the modified CIECAM02 can predict
smartphones accurately, which appear to be satisfactory.
6.1 iPhone data analysis
The data analysis and comparison of iPhone experiments was carried out to
understand the colour reproduction consistency between them by adding three
more iPhones to this stage of the study and also investigating iPhone colour
reproduction consistency mainly focused on cyan colours, with setting up 10
more new samples to the experiment. Referring to the first step of the iPhone
colour experiment, there was a scattering of hue points mainly in the cyan colour
which was worth more investigation. Following further experimentation, no
significant change was recorded therefore the reason of scattering in the hue
chart remains inconclusive and so must be put down to a subject reading error or
recording error .
Elham Khodamoradi Page 129
iPhone Lightness graphs
Figure 6.2: The prediction of iPhone lightness. [72]
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
3
Mean
Lightness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
6
Mean
Lightness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
2
Mean
Lightness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
1
Mean
Lightness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
3
Mean
Lightness-iphone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
4
Mean
Lightness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
5
Mean
Lightness-iPhone
Elham Khodamoradi Page 130
Figure 6.3: iPhone lightness prediction for cyan samples. [73]
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
9
Mean
Lightness _iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
8
Mean
Lightness _iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
7
Mean
Lightness _iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
6
Mean
Lightness-iphone
Elham Khodamoradi Page 131
Figure 6.4: The prediction of iPhone colourfulness. [74]
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
1
Mean
Colorfulness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
2
Mean
Colorfulness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
3
Mean
Colorfulness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
4
Mean
Colorfulness-iPhone
0
20
40
60
80
100
120
0 20 40 60 80 100
sub
ject
5
Mean
Colorfulness-iPhone
0
20
40
60
80
100
0 20 40 60 80 100
sub
ject
6
Mean
Colorfulness-iPhone
Elham Khodamoradi Page 132
Figure 6.5: iPhone colourfulness prediction for cyan samples. [75]
0
20
40
60
80
100
0 20 40 60 80 100
Me
an s
ub
ject
1
Mean
Colorfulness
0
20
40
60
80
100
0 20 40 60 80 100
Me
an s
ub
ject
2
Mean
Colorfulness
0
20
40
60
80
100
0 20 40 60 80 100
Me
an s
ub
ject
3
Mean
Colorfulness
Elham Khodamoradi Page 133
Figure 6.6: The prediction of iPhone hue. [76]
0
100
200
300
400
0 100 200 300 400
sub
ject
1
Mean
Hue-iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
2
Mean
Hue-iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
3
Mean
Hue-iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
4
Mean
Hue-iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
5
Mean
Hue-iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
6
Mean
Hue-iPhone
Elham Khodamoradi Page 134
Figure 6.7: iPhone hue prediction for cyan samples. [77]
Experiments 1 to 9 are performed on colour samples by iPhone as illustrated in
Figure 6.8 below, with a graphical comparison of the variability between the sets
0
100
200
300
400
0 100 200 300 400
sub
ject
1
Mean
Hue _iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
2
Mean
Hue _iPhone
0
100
200
300
400
0 100 200 300 400
sub
ject
3
Mean
Hue _iPhone
Elham Khodamoradi Page 135
of samples in terms of the three attributes through comparison between
Experiment 1 (x-axis) and the other 8 experiments.
0
20
40
60
80
100
120
0 20 40 60 80 100
Me
an L
igh
tne
ss
Mean
Lightness
experiment 1
experiment 2
experiment 3
experiment 4
experiment 5
experiment 6
experiment 7
experiment 8
experiment 9
0
20
40
60
80
100
0 20 40 60 80 100
Me
an C
olo
rfu
lne
ss
Mean
Colorfulness
experiment 1
experiment 2
experiment 3
experiment 4
experiment 5
experiment 6
experiment 7
experiment 8
experiment 9
Elham Khodamoradi Page 136
Figure 6.8: Comparison results between mean estimations of lightness (top), colourfulness (middle) and hue (bottom) for iPhone Experiment 1 (x-axis) and Experiments 2-9 (y-axis). [78]
0
100
200
300
400
0 100 200 300 400
Hu
e L
igh
tne
ss
Mean
Hue
experiment 1
experiment 2
experiment 3
experiment 4
experiment 5
experiment 6
experiment 7
experimeny 8
experiment 9
Elham Khodamoradi Page 137
7. Results and Discussions
7.1 Testing CIECAM02
The CIECAM02 model is the colour appearance model recommended by CIE.
The model’s performance is evaluated using the present experimental data in
terms of CV calculated between the model’s predicted and visual results.
The results shows that the model gave a reasonable prediction for the brightness
and hue visual results, but performed poorly for predicting colourfulness visual
results, i.e. the correlation coefficient (r) for CIELUV L* are 0.963, 0.959, 0.960,
and 0.940 for the iPhone, LG Nexus4, HuaWei and SamSung respectively, and
are 0.890, 0.876, 0.761, and 0.764 respectively for CIELUV C* when comparing
with the counterparts on the CRT monitor.
To predict a colour on a mobile telephone using CIECAM02, the predictions rest
on a number of environmental parameters settings, e.g., f = 0.9, c = 0.59, and
nc = 0.90, which gives closer results with less scattering.
Figure 5.6 presented a plot of a hue-colourfulness with regard to hue
estimations, although the correlation for hue remains the highest (i.e. r = 0.96)
between the iPhone and CRT, i.e. greenish colours appear to be greener and
blue colours bluer as shown in Figure 5.15. All colours tended to be more
colourful when depicted on the phone.
Figure 5.16 shows the plots for lightness, colourfulness and hue visual results
against the corresponding CIECAM02 predictions, respectively.
Elham Khodamoradi Page 138
Figure 6.1 shows that by modifying CIECAM02, we can predict smartphone
colours accurately, which appear to be satisfactory when viewed.
Figure 8.1 shows a comparison of the iPhone (x-axis) with mobile telephones
form LG-Nexus4 (top), Samsung (middle) and Huawei (bottom) by modified
CIECAM02. When comparing with the other smartphones, in terms of hue, all
three types of phones tend to be more reddish for purplish colours than those
displayed on an iPhone, whereas the rest remains nearly the same. With regard
to lightness, for lighter colours, all three telephones unanimously appear darker
than on iPhones with the LG-Nexus being the darkest, 25% darker, whereas
18% and 16% darker are evidenced for Huawei and Samsung respectively.
However, the opposite phenomenon occurs when it comes to the representation
of colourfulness. The images on all three telephones, the LG-Nexus4, Samsung,
and Huawei, appear more colourful than those displayed on an iPhone. [Park,
2006].
7.2 Refinement of CIECAM02
In the last section, it was found that the performance of CIECAM02 is not
completely satisfactory, i.e. the results in terms of CV in predicting current results
are much better than those in predicting smartphones data. Hence, new trials
were made to try to improve the model’s performance for predicting visual data.
The general strategy was to add 10 more samples to the whole experiment and
use 3 more iPhones in order to repeat the experiment and mainly focus on 10
new sample slides which had been carefully chosen from cyan and blueish
colours as by analysing the results, in the iPhone there is a slight confusion
around cyan colour. Comparison between the new set of samples and the old
Elham Khodamoradi Page 139
samples has also been taken into account in order to have a better and clearer
understanding of the issues so far.
Table 7.1 below summarises the values for the iPhone based on experimental
data obtained from subject’s readings.
Table 7.1: Values for iPhone Experiment. [10]
Subject
Lightness Colourfulness Hue
r a b r a b r a b
A 0.84 1.00 -23.5 0.94 1.21 -18.7 0.75 0.97 30
B 0.88 1.00 -13.95 0.92 1.21 -16.6 0.74 0.94 39.3
C 0.88 1.14 -24.3 0.84 0.73 20.3 0.72 1.01 -10.7
D 0.82 0.93 -11.6 0.85 0.88 12.65 0.8 1.1 -8.1
E 0.93 1.04 -14.25 0.57 0.75 24.35 0.66 1.03 -9.1
F 0.88 1.00 -13.95 0.71 0.77 14.64 0.55 0.86 30.5
Mean 0.87 1.01 -13 0.80 0.92 12.34 0.70 0.98 11.9
As shown in Table 7.1, each estimate by the subject is measured using three
values,𝑎, 𝑏 and 𝑟 (correlation coefficient).
Elham Khodamoradi Page 140
8. Modelling of iPhone appearance using CIECAM02
Since the colour appearance model CIECAM02 was developed for the medium
of a CRT, which features refection and transparency, it is not well equipped to
predict smartphones. After setting the environmental parameters to ‘dim’
conditions where F = 0.9, c = 0.59, and Nc = 0.90 to compensate for lightness
differences between a CRT and iPhone, the comparison results are given in
Figure 6.1.
The general strategy was to modify the CIECAM02 model as little as possible.
The modification was made for the three iPhones, where colourfulness was
adjusted according to equation below:
Colourfulness_smartphone = 1.8 * Colourfulness_CIECAM02
The new equations were used to replace the fixed parameters used by
CIECAM02 for the iPhone experiment. Its predictions and visual results are
plotted in Figures 6.1 for lightness, colourfulness and hue results respectively.
After correction using the equation above, the modified CIECAM02 can predict
smartphones accurately.
Figure 6.1 shows that there is a good prediction by the refined CIECAM02. It
can be seen clearly in Figures 6.1 that the largest improvement can be found in
colourfulness, followed by lightness and hue the smallest.
Elham Khodamoradi Page 141
8.1 Comparison with other smartphones
After the modelling of an iPhone using CIECAM02, a number of other
smartphones can be evaluated as well in terms of colour appearance, including a
LG-Nexus4, Samsung, and Huawei.
Figure 8.1 presents the comparison results from the calculations using
CIECAM02 for all the telephones when the settings are given as F = 0.9, c =
0.59, and Nc = 0.90 and the colourfulness applies Eq. (8.1).
Figure 8.1: Comparison of iPhone (x-axis) with phones of LG-Nexus4 (top), Samsung (middle)
and Huawei (bottom) by modified CIECAM02. [80]
When comparing with the other smartphones, in terms of hue, all three types of
telephone tend to be more reddish for purplish colours than those displayed on
an iPhone, whereas the rest maintains near the same. With regard to lightness,
Elham Khodamoradi Page 142
for lighter colours, all three telephones unanimously appear darker than on
iPhones with the LG-Nexus being the darkest, 25% darker, whereas 18% and
16% darker are evidenced for HuaWei and SamSung respectively. However, the
opposite phenomenon occurs when it comes to the representation of
colourfulness. The images on all three telephones, the LG-Nexus4, SamSung,
and HuaWei, appear more colourful than those displayed on an iPhone. For
example, the colours on a LG-Nexus4 telephone appear systemically 10% more
colourful than those depicted on an iPhone. In addition, for both the HuaWei and
SamSung phones, they appear again to be 17% and 22% more colourful,
especially for colourful samples, the tendency presenting across all three
telephones. Since these findings are based on only one telephone of each type,
future study is needed to focus on the investigation of large samples with more
similar phones.
Figure 8.2 demonstrates the colour check images that are displayed on each
mobile telephone together with one from a CRT. As indicated by arrows, the
purple colour appears to be slightly different on different telephones.
Elham Khodamoradi Page 143
Figure 8.2: Colour checker depicted using four phones and a CRT. [81]
Elham Khodamoradi Page 144
9. Summary of Phase 2
With 85% of smartphone users performing their everyday tasks using the device
due to its ability and connectivity, the smartphone continues to revolutionize how
people perform their everyday activities. It is expected that this work will
contribute to this revolution and complement users with some information when
they perform online shopping buying colour sensitive goods.
In summary, in terms of the measurement on each mobile telephone,
encouragingly, the variations among the same kind of phones are insignificant,
being less than 3 E CIELAB units.
In addition, when comparing with subjects’ hue estimations, all phones and the
CRT monitor appear to have similar hue values, indicating that the hue values
have been well preserved on those phones although some variations are
observed. However, when it is viewed on a mobile telephone, a colour appears
to show a variance with colourfulness by appearing much more colourful.
Furthermore, the correlation coefficients (r) for CIELUV L* are 0.963, 0.959,
0.960, and 0.940 for the iPhone, LG Nexus4, HuaWei and SamSung
respectively, and are 0.890, 0.876, 0.761, and 0.764 respectively for CIELUV C*
when comparing with the counterparts on the CRT monitor. Therefore, the
iPhone tends to be the best with fewer scatterings. To predict a colour on a
mobile telephone using CIECAM02, the predictions rest on a number of
environmental parameter settings, e.g. f = 0.9, c = 0.59, and nc = 0.90, which
gives closer results with less scattering. Consequently, for an image with truthful
colour to be displayed on a mobile phone, the forward and reverse model of
Elham Khodamoradi Page 145
CIECAM02 will have to be applied. Specifically, for the iPhone 5, nearly half of
the colourfulness predicted by CIECAM02 needs to be factored in. Further
studies are in place to take a bigger number of mobile telephones into
consideration.
Elham Khodamoradi Page 146
10. Conclusion and future work
The colour appearance model CIECAM02 was standardised in 2002 and has
since been widely applied to predict colour appearance under varying viewing
environments. However, it does not cover mobile devices that have become
available and popular in the last decade or so. To fill the gap, this research
aimed to ensure the CIECAM02 model can predict colours on mobile telephones.
Also, the study of textured colours was carried out in order to fill another gap that
CIECAM02 is missing.
For texture colours, liner modification appears to be adequate since textured
colours appear to be darker and more colourful. However, only 5 texture patterns
are studied in this research, more experiments should be designed in the future
to further understand the effect of texture on human perception.
10.1 Measuring the Impact of Texture to Colour Appearance model using
CIECAM02
Research into colour measurement using a calibrated colour imaging systems is
investigated by measuring and analysing the effect of texture on colour
appearance. The analysis is based on measured colour and the CIE co-ordinate
colour definitions of coloured image phenomena. Colour phenomena are used to
define the variation in colour appearance of different textures introduced in this
study.
The study is designed to establish as separate variables of colour appearance, a
model of colorant formulations, and a model of the colour appearance of that
formulation when applied to a given texture. The CIECAM02 model was
Elham Khodamoradi Page 147
validated using a visual simulation of colour appearance on a computer screen
that has been calibrated so that each colour phenomena could be controlled and
reproduced to within an average D65 CIE 10 degree Observer colour difference.
Colour textured samples with a range of 6 different textures have been used in
all the experiments in this research. They focus on shapes such as dots, lines
and crosses, etc. The measuring and analysing properties of Tristimulus values
leads to the modelling of the relationship between variations in colour
appearance due to texture, and the fundamental colour represented by a
colorant formulation. The study looks to establish computational relationships
between colour and texture as distinct contributing variables of colour
appearance.
The results of visual matching experiments (Experimental 2-6) are the basis for
colour modelling, and have been used to define the psychophysical attributes of
the human sensation of colour, such as lightness, colourfulness and the hue of
the colour as well as the modification of the colour model by using texture and
comparing the interaction of it in the colour model. Both the colorimetric accuracy
of the data derived from images, and the accuracy of the matching process itself
is assessed.
Surface texture effects are responsible for significant visual colour shifts as is
shown in the results. The variability of such observer trials requires output to be
generated from multiple observations (See Results tables 1 & 2), and it is
important for the results of on-screen matching experiments to be subjected to
statistical analysis for significance. In the following discussion, some results are
reported for the effects of texture on colour appearance. In a separate series of
Elham Khodamoradi Page 148
experiments [2-6], multiple observations were carried out in a variety of matching
tasks, ranging from the simple texture such as dots and line to more complex
texture in the samples under viewing cabinet conditions. Full details and the
calculation are reported separately in the appendix, and the results are
summarized here for reference.
From Table 4.2, the reasonable mean values can be seen. Experiment 4 and, to
a lesser extent Experiment 6, were reported as ‘difficult to measure’ as they had
a more complex texture present. By looking at the Mean and CV values of the
experiments, the appearance of the texture will move more smoothly through
colour-space. Table 4.2 indicates that the presence of texture in the simulation,
has some effects on an observer’s measurements.
10.2 Analysis of more complex Texture effects From Experiment 4 results it can be seen that the collected data are not
consistent with other performed texture-related experiments. The complex
texture is viewed and quantified as consistently less colourful in all 30 samples,
compared to the other textured experiments. It viewed as more colourful than all
cases and more colourful than Experiment 6 as shown in Figure 4.8 that
represents the higher colourfulness reading.
As can be seen from Figure 4.8, the largest discrepancies occur on colourfulness
estimations, in particular for Experiments 4 () and 6 (), which contain patterns
of random lines and random dots as depicted in Figure 4.2, which could be
explained on the basis that random patterns make colourfulness more difficult to
estimate. Although Experiment 2 also contains patterns scattered around the
colour sample, the texture pattern itself, i.e. X, is symmetrical. Otherwise, all the
Elham Khodamoradi Page 149
points in Figure 4.8 are slightly above the central line for the colourfulness
estimation, suggesting that texture has some effect on colourfulness estimations
and makes colours appear to be more colourful.
For lightness, the textured colours appear to be darker than the colours without
textures, which indicates that the texture decreases the lightness contrast and
makes colours appear darker. On the other hand, little effect is found for hue
estimation, which is expected as the texture patterns are neutral colour, the
same grey as the background. Therefore it is concluded in this study that texture
that shares the same grey level with that in the background has little effect on
hue estimation, makes a colour appear more colourful and darker, while under
the D65 viewing condition on CRT monitors.
The primary application of phase 1 of this study was to calculate colour
differences between two sets of experiments and study six different texture
patterns under the same backgrounds. In the future it would be ideal to perform
content-based image retrieval on a variety of patterns overlaid on various
coloured backgrounds (e.g. a collection of wallpaper images
http://www.moda.mdx.ac.uk/collections/wallpapers) as extensive investigation
has been carried out in this study only on a grey background with simple textured
pattern.
Phase 2 of this study introduces measures of colour appearance in smartphones
based on visual matching methods. Colour based colorimetric analysis was
facilitated in these experiments, by the availability of measuring the samples on
the smartphone screen under controlled viewing conditions. Images have been
classified according to the colour appearance descriptions which humans
Elham Khodamoradi Page 150
understand. In the future this will allow people to find coloured images more
easily on devices. Overall, the metrics developed using in the CIECAM02 colour
model, to mark-up images with colour appearance concepts, show positive
results on smartphones and more strongly on iPhones.
In conclusion for mobile telephones, it appears that colours appear lighter and
more colourful than on CRT monitors. However, only a limited number of mobile
telephones were tested i.e. 6 iPhone5s and one Samsung, Huawei and LG
Nexus, these conclusions might be limited. In the future, it is planned that more
mobile telephones are to be studied to confirm the results achieved in this work.
In particular, the iPhone5 seems to be prevalent, especially among those
observers who helped doing the experiments.
The results from the proposed methods show that this is a contribution to
bridging the semantic gap in the area of whole-scene colour appearance in
smartphones as they now become a daily use machine in a wide range of
procedures from watching a movie to purchasing items online.
Initial small-scale human observation tests have been carried out and extensive
tests are planned for future developments such as using more complex sample
texts with patterns. Subjective trials are important to determine quality of the
algorithms and classifications. These colour appearance classifications can be
integrated into a smartphone image browsing system which allows them to also
be used to refine browsing. For example, in browsing certain collections, e.g.
multi-coloured items such as a mug with bright and intense colours, or to search
for some single colour chair or shoes. These descriptors can therefore help the
searching process. Finally, in the future, it would be ideal to plan to use this
Elham Khodamoradi Page 151
modification, with the aim that the analysis will take into account the colour
appearance more accurately in smartphones particularly in popular devices such
as the iPhone.
Elham Khodamoradi Page 152
Appendixes:
Appendix A: The calculation of CIECAM02
Appendix B: The comparison of each individual’s estimations with the mean
results for Experiment 1 in Phase 1.
Appendix C: The CV values of 7 subjects for performing six experiments in
Phase 1.
Elham Khodamoradi Page 153
Appendix A: The calculation of CIECAM02
1. Firstly, the measurement using a colour meter for luminance, reference white,
background, test colour samples take place to obtain LA, LA-bg, and X, Y, Z values.
2. View conditions and notations where is a newly introduced factor in the CAMcc model
for the calculation of simultaneous contrast.
Surround F C Nc
Average 1.0 0.69 1.0
Dim 0.9 0.59 0.95
Dark 0.8 0.525 0.8
Luminous computer monitor 0.2 0.41 0.80
LA Luminance of reference white in cd/m2
LA-bg Luminance of background in cd/m2
Yb Y value for background ranging within [1,100].
Yw Y value for reference white and close to 100.
𝑘 =1
5𝐿𝐴 + 1
𝐹𝐿 = 0.2𝑘4(5𝐿𝐴) + 0.1(1 − 𝑘4)2(5𝐿𝐴)1/3
𝑛 =𝑌𝑏
𝑤
𝑁𝑏𝑏 = 𝑁𝑐𝑏 = 0.725(1
𝑛)0.2
𝑧 = 1.48 + √𝑛
Elham Khodamoradi Page 154
3. Chromatic adaption
[𝑅𝐺𝐵
] = 𝑀𝐶𝐴𝑇02 [𝑋𝑌𝑍
] (A.1)
𝑀𝐶𝐴𝑇02 = [0.7328 0.4296 0.16240.7036 1.6975 0.00610.0030 0.0136 0.9834
] (A.2)
(A.3)
𝑅𝑐 = 𝑅 [𝐷𝑌𝑤
𝑅𝑤+ 1 − 𝐷] (A.4)
𝐺𝑐 = 𝐺 [𝐷𝑌𝑤
𝐺𝑤+ 1 − 𝐷] (A.5)
𝐵𝑐 = 𝐵 [𝐷𝑌𝑤
𝐵𝑤+ 1 − 𝐷] (A.6)
[𝑅′
𝐺′
𝐵′] = 𝑀𝐻𝑀𝐶𝐴𝑇02
−1 [
𝑅𝑐
𝐺𝑐
𝐵𝑐
] (A.7)
𝑀𝐶𝐴𝑇02−1 = [
1.0961 0.2788 0.18270.4543 0.4735 0.07200.0009 0.0056 1.0153
] (A.8)
𝑀𝐻 = [0.3897 0.6889 0.07860.2298 1.1834 0.04640.0000 0.0000 1.0000
] (A.9)
4. Non-linear Response Compression
𝑅𝑎′ =
400(𝐹𝐿𝑅′
100)
0.42
27.13+(𝐹𝐿𝑅′
100)
0.42 + 0.1 (A.10)
𝐺𝑎′ =
400(𝐹𝐿𝐺′
100)
0.42
27.13+(𝐹𝐿𝐺′
100)
0.42 + 0.1 (A.11)
Elham Khodamoradi Page 155
𝐵𝑎′ =
400(𝐹𝐿𝐵
100)
0.42
27.13+(𝐹𝐿𝐵′
100)
0.42 + 0.1 (A.12)
5. Perceptual attribute correlates
𝑎 = 𝑅𝑎′ −
12𝐺𝑎′
11+
𝐵𝑎′
11 (A.13)
𝑏 =1
9(𝑅𝑎
′ + 𝐺𝑎′ − 2𝐵𝑎
′ ) (A.14)
Hue angle:
ℎ = 𝑡𝑎𝑛−1(𝑏
𝑎) (A.15)
Eccentricity factor:
𝑒𝑡 = [12500
13𝑁𝑐𝑁𝑐𝑏] [cos (ℎ
𝜋
180+ 2) + 3.8] (A.16)
𝑡 =50(𝑎2+𝑏2)
12100𝑒𝑡(
10
13)𝑁𝑐𝑁𝑐𝑏
𝑅𝑎′ +𝐺𝑎
′ +21
20𝐵𝑎
′ (A.17)
Hue response:
𝐻 = 𝐻𝑖 +
100(ℎ−ℎ𝑖)
𝑒𝑖ℎ−ℎ𝑖
𝑒𝑖+
ℎ𝑖+1−ℎ
𝑒𝑖+1
(A.18)
where ℎ𝑖 ≤ ℎ < ℎ𝑖+1 and if ℎ > ℎ5, ℎ = ℎ − 360.
Elham Khodamoradi Page 156
Red Yellow Green Blue Red
i 1 2 3 4 5
hi 20.14 90 164.25 237.53 380.14
ei 0.8 0.7 1.0 1.2 0.8
Hi 0 100 200 300 400
Achromatic Response:
𝐴 = [2𝑅𝑎′ + 𝐺𝑎
′ + (1
20) 𝐵𝑎
′ − 0.305] 𝑁𝑏𝑏 (A.19)
Lightness:
𝐽 = 100(𝐴
𝐴𝑤)𝑐𝑧 (A.20)
where Aw is the A value for reference white.
Brightness:
𝑄 =4
𝑐(
𝐽
100)0.5(𝐴𝑤 + 4)𝐹𝐿
0.25 (A.21)
where Aw is the A value for reference white.
Chroma:
𝐶 = 𝑡0.9(𝐽
100)0.5(1.64 − 0.29𝑛)0.73 (A.22)
Colourfulness:
𝑀 = 𝐶𝐹𝐿1/4 (A.23)
Saturation:
𝑠 = 100√𝑀
𝑄 (A.24)
Elham Khodamoradi Page 157
Appendix B: The comparison of individual estimation with the mean results for Experiment 1 in Phase 1.
Elham Khodamoradi Page 158
Elham Khodamoradi Page 159
Appendix C: The CV values of 7 subjects for performing six experiments in
Phase 1.
Table C-1: The CV values of all seven subjects who performed Experiment 1
(Colour Sample).
Subject
Lightness Colourfulness Hue
CV r a b CV r a b CV r a b
A 20 0.88 1.05 -2.7 32 0.62 1.6 1.15 9 0.95 1.04 -17.7
B 17 0.87 0.76 11.8 26 0.44 0.54 18.23 11 0.94 1.06 -15.6
C 36 0.89 1.18 -21.9 26 0.61 0.789 4.05 13 0.92 1.08 15.67
D 41 0.89 1.14 -11.2 20 0.81 1.14 -6.72 9 0.96 0.95 12.92
E 29 0.62 0.58 29.78 20 0.64 1.06 7.94 14 0.93 1.02 -11.7
F 25 0.91 1.26 -3.21 16 0.64 0.76 6.83 16 0.95 1.07 -34.3
G 23 0.9 1.18 -10.3 63 0.60 0.779 25.66 9 0.95 1.06 2.3
Mean 27 0.82 1 0.003 29 0.62 0.902 12.72 11 0.90 1.466 0.01
Elham Khodamoradi Page 160
Table C-2. CV values for Experiment 2 (Textured Colour Sample)
Subject
Lightness Colourfulness Hue
CV r a b CV r a b C
V r a b
A 24 0.94 1.14 -15.4 29 0.72 1.00 -1.04 11 0.74 0.91 -4.74
B 17 0.95 1.13 -9.52 31 0.79 1.1 -7.65 11 0.81 1.05 -24.2
C 14 0.93 0.9 3.43 20 0.88 1.06 -4.77 14 0.89 1.19 -16.3
D 24 0.84 1.05 -3.44 23 0.79 0.92 -0.96 13 0.7 0.85 37.2
E 27 0.74 0.73 12.4 27 0.77 0.94 5.66 14 0.91 1.23 -51.2
F 23 0.88 1.08 3.38 43 0.74 0.93 16.5 15 0.74 1.94 33.9
G 22 0.9 0.94 9.09 43 0.7 0.68 24.7 18 0.74 0.87 25.4
Mean 21 0.88 30 0.77 13
Elham Khodamoradi Page 161
Table C-3. CV values for Experiment 3 (Textured Colour Sample)
Subject
Lightness Colourfulness Hue
CV r a b CV r a b C
V r a b
A 38 0.51 0.52 40.9 49 0.76 0.89 1.17 11 0.67 0.92 26.2
B 22 0.93 1.12 -6.08 30 0.83 1.09 -8.11 11 0.71 0.89 38.2
C 27 0.87 1.19 -17.3 28 0.78 0.85 4.59 7 0.71 0.95 11.4
D 53 0.94 1.16 -7.88 40 0.83 1.09 0.04 19 0.85 1.16 -61.4
E 32 0.94 1.16 -7.88 34 0.59 0.74 15 11 0.85 1.18 -40.6
F 12 0.77 0.83 0.42 21 0.8 0.95 -0.7 8 0.74 0.91 27.5
G 29 0.92 1.1 -6.37 29 0.77 1.15 14.2 16 0.78 0.96 -1.16
Mean 30 33 11
Elham Khodamoradi Page 162
Table C-4. Experiment 4 (Textured Colour Sample)
Subject
Lightness Colourfulness Hue
CV r a b CV r a b C
V r a b
A 20 0.89 1.36 -27.9 32 0.89 1.31 -11.1 9 0.86 0.98 -7.71
B 17 0.81 0.92 3.41 26 0.87 1.01 5.84 11 0.9 1.05 -17.6
C 36 0.92 1.08 -5.37 26 0.93 1.2 -7.05 13 0.92 1.11 -24.7
D 41 0.75 0.84 17.8 20 0.72 0.78 14.7 9 0.68 0.74 76.7
E 29 0.84 0.94 -2.27 20 0.82 0.69 6.52 14 0.84 0.99 -6.66
F 25 0.91 1.05 -2.61 16 0.93 1.11 -4.49 16 0.91 1.09 -17.0
G 23 0.85 0.79 16.9 28 0.9 0.86 5.59 9 0.87 1.01 -3.78
Mean 27 24 11
Elham Khodamoradi Page 163
Table C-5. Experiment 5 (Textured Colours Sample)
Subject
Lightness Colourfulness Hue
CV r a b CV r a b C
V r a b
A 26 0.94 1.25 -20.5 26 0.83 1.21 -4.87 18 0.75 0.94 6.31
B 21 0.87 0.9 5.34 38 0.59 0.72 20.3 24 0.57 0.67 112.
C 20 0.88 0.95 3.59 24 0.84 1.13 -6.01 12 0.9 1.15 -41.2
D 27 0.93 1.17 -17.5 22 0.86 1.14 -1.79 12 0.89 1.13 -34.7
E 37 0.79 0.89 14.7 30 0.76 1.02 -4 16 0.82 1.02 -1.29
F 18 0.91 0.93 6.92 32 0.65 0.79 13.2 12 0.88 1.04 -31.5
G 19 0.9 0.87 7.56 29 0.62 0.7 12.6 14 0.88 1.02 -10.2
Mean 24 28 15
Elham Khodamoradi Page 164
Table C-6. Experiment 6 (Textured Colours Sample)
Subject
Lightness Colourfulness Hue
CV r a b CV r a b C
V r a b
A 17 0.96 1.02 0.97 21 0.88 0.9 2.04 15 0.78 0.99 11.8
B 16 0.97 1.02 1.85 20 0.88 0.93 -0.54 15 0.78 0.98 14.2
C 17 0.98 1.21 -11.5 35 0.86 1.2 -23.2 11 0.84 1.05 -5.97
D 17 0.96 1.1 -3.92 24 0.89 1.25 -17.1 10 0.87 1.11 -32.1
E 12 0.97 1.08 -5.58 30 0.89 1.2 -0.9 13 0.82 1.03 -16
F 29 0.86 0.67 17.5 31 0.73 0.68 23.9 14 0.82 0.96 -11.0
G 22 0.93 0.78 11.0 26 0.78 0.82 15.9 19 0.67 0.84 39.0
Mean 18 26 13
Elham Khodamoradi Page 165
Appendix D: The Paper published at AIC 2015, Tokyo, May 2015.
Evaluation of colour appearances displaying on
smartphones
X. Gao 1*, E. Khodamoradi 1, L. Guo 2, X. Yang 2, S. Tang 3, W. Guo 3, Y. Wang 3
1 Department of Computer Science, Middlesex University, London, NW4 4BT,
UK
2 Department of Information Science, Fuzhou University, Fuzhou, China
3 Biomedical Engineering Center, Fudan University, Shanghai, China
*Corresponding Author: Prof. Xiaohong Gao, [email protected]
ABSTRACT
Despite of the limited size and capacity of a mobile phone, the urge to apply it to
meet quotidian needs has never been unencumbered due to its appealing
appearance, versatility, and readiness, such as viewing/taking pictures and
shopping online. While a smartphone can act as a mini-computer, it does not
always offer the same functionality as a desktop computer does. For example,
the RGB values on a smartphone normally cannot be modified nor can white
balance be checked. As a result, performing online shopping using a mobile
phone can be tricky, especially when buying colour sensitive items. Therefore,
this research takes an initiative to investigate the variations of colours for a
Elham Khodamoradi Page 166
number of smartphones while making an effort to predict their colour appearance
using CIECAM02, benefiting both phone users and makers. The paper studies
models of Apple iPhone5, LG Nexus 4, Samsung, and Huawei, by capitalising on
comparisons with a CRT colour monitor that has been calibrated under the
illuminant of D65, to be in keeping with the usual way of viewing online colours.
As expe cted, all the phones present more colourful images than a CRT does.
Elham Khodamoradi Page 167
1. Introduction
A smartphone is a mobile phone with more advanced computing capability and
connectivity than basic feature phones, offering functionalities of
typically personal assistance, media player, digital camera, and a GPS
navigation unit in addition to the basic calling/receiving facilities. At present, the
global smartphone audience has reached 1 billion consumers [eMarkter] and
expects to arrive at 1.75 billion by the end of 2014. As such, mobile sales are
not only focusing heavily on smartphones, but also on the more affordable option
of feature phones that do not have an operating system. As a result, mobile
penetration has surpassed 100% in many regions of the world, including North
America, Western Europe, Central and South America, Central and Eastern
Europe, and the Middle East [WeAreSocial]. Among those smartphone users,
about half of them go online by using the phone regularly. Table 1 lists the top 10
most popular smartphones on the market.
Table 1. Top 10 smartphone list as of May 2014 [Counter].
Rank Brand Model
1 Apple iPhone 5s
2 Samsung Galaxy S5
3 Samsung Galaxy S4
4 Samsung Note 3
5 Apple iPhone 5c
6 Apple iPhone 4S
7 Xiaomi MI3
8 Samsung Galaxy S4
mini
Elham Khodamoradi Page 168
One of the unexpected by-products of smartphones remains in the field of digital
photography. It appears that more photos are taken by using Smartphones than
normal cameras. For example, the Apple iPhone 5 is currently the most popular
'camera' on Flickr.com [Flickr]. As a direct result, large amount of money are
being investigated by mobile developers to create photo apps in an attempt to
satisfy the demands of 'serious' camera phone photographers.
While using smartphone to perform everyday activities, colour remains one of the
key factors, in particular in online shopping. Similar to any other digital device, a
phone represents a digital image in a RGB colour space. Therefore when an
image is to be processed, it is usually firstly converted into the colour space of,
say, hue, lightness and colourfulness. In this way, the dependency of RGB space
on hardware devices can be circumvented, i.e., a colour in one device usually
does not appear nor measure the same as the one in another device even with
the same RGB values in both devices. This is because the range of R, G, or B
values are manually set to be the same (such as [0, 255] for an 8-bit computer)
for all devices regardless their physical measurements. On the other hand, hue,
colourfulness and lightness, space agrees more with human vision theories. To
further improve the fitness between users’ perception and retrieved results, CIE
has recommended a colour appearance model CIECAM02 to predict colours
appear on any media under a number of viewing conditions [Moroney]. Stemmed
9 Xiaomi Hongmi
Redrice
10 Samsung Galaxy
Grand 2
Elham Khodamoradi Page 169
from Hunt’s early colour vision model [Luo (a) 1993, Luo (b) 1993 ] employing a
simplified theory of colour vision for chromatic adaptation together with a
uniformed colour space, CIECAM02 can predict the change of colour
appearance as accurately as an average observer under a number of given
viewing conditions. In particular, the way that the model describes a colour is
reminiscent of subjective psychophysical terms, i.e., hue, colourfulness, Chroma,
brightness and lightness.
To begin with, CIECAM02 takes into account of measured physical parameters
of viewing conditions, including tristimulus values (X, Y, and Z) of a stimulus, its
background, its surround, the adapting stimulus, the luminance level, and other
factors such as cognitive discounting of the illuminant. The output of the colour
appearance model predicts mathematical correlates of perceptual attributes.
With regard to the representation of the colour appearance of an image, in this
investigation, the perceptual colour attributes of lightness (J), colourfulness (M)
and hue (H) are employed.
Elham Khodamoradi Page 170
2. Method
Thirty test colours are randomly selected from the Munsell colour book while
making an effort to cover as much CIE 1931 colour space as possible.
Psychophysical experiments are then carried out on both 19” CRT colour monitor
with its illuminant calibrated to D65 and mobile phones. As illustrated in Figure
1, each test colour is placed at a centre against a grey background (with 20% of
luminance of reference white) and surrounded by the reference white, reference
colourfulness and surrounding colours. The test field in the centre subtends a
visual angle of 2o at a viewing distance of ~60cm. Ten subjects with normal
colour vision are selected to conduct the experiments using the technique of
magnitude estimation which they have been trained in advance to apply skilfully.
Specifically, for each test colour, each subject is asked to estimate its
appearance in terms of lightness, colourfulness and hue contents verbally that
are then recorded by an operator sitting nearby.
Figure 1. The paradigm of the experimental pattern.
Figure 2. The x, y data of the thirty sample colours plotted on the CIE xy Chromaticity diagram for the four phones to be studied in this research. The big * refers to the reference white.
Elham Khodamoradi Page 171
In addition, each colour on each phone has been measured using a colour meter
CS-100A, simulating subjects’ viewing position. Specifically, the reference white,
reference colourfulness, and background are measured at least 3 times, e.g., at
the beginning, the middle and the end of the colour sample sequence, to check
the repeatability of the phone. In total, 6 phones, including three iPhone5, one
Huawei, one SamSung, and one LG Nexus4, are measured and estimated. The
same work is also performed on the CRT monitor, Philips Brilliance 201B. The
measured data are presented in a CIE xy Chromaticity diagram as illustrated in
Figure 2.
Elham Khodamoradi Page 172
3. Results and Discussion
3.1 iPhone
Three handsets of iPhone 5 are investigated in this paper. Figure 2 compares
the subjects’ estimation results between CRT and iPhone 5 where correlation
coefficient (r) values are 0.76, 0.62 and 0.96 respectively for Lightness,
colourfulness, and hue estimations.
For lightness, the estimation on mobile iPhone tends to be 16% more than that
on CRT monitors, whereas 11% increase of colourfulness for mobile phones is
evidenced. In Figure 4, a hue-colourfulness plot is presented, where small circles
(o) represent the colours from CRT and big square (□) from iPhone5.
Figure 3. Comparison of subjective estimations of the thirty samples for CRT (x-axis) and iPhone 5 (y-axis).
Figure 4. Subjects’ estimation for the same group of colours on both CRT and Iphone5. Small circle(o) refers to the estimation for CRT whereas big solid square (▪) on Iphone.
3.2 Modelling of Iphone appearance using CIECAM02
Since the colour appearance model CIECAM02 is developed for the media of
CRT, refection and transparency, it may not be well equipped to predict
Elham Khodamoradi Page 173
smartphones. After the setting of environmental parameters to ‘dim’ condition
where F =0.9, c = 0.59, and Nc=0.90 to compensate lightness differences
between CRT and iPhone, the comparison results are given in Figure 6 for the
three phones, where colourfulness is adjusted according to Eq. (1).
𝐶𝑜𝑙𝑜𝑢𝑟𝑓𝑢𝑙𝑛𝑒𝑠𝑠_𝑠𝑚𝑎𝑟𝑡𝑝ℎ𝑜𝑛𝑒 = 1.8 ∗ 𝐶𝑜𝑙𝑜𝑢𝑟𝑓𝑢𝑙𝑛𝑒𝑠𝑠_𝐶𝐼𝐸𝐶𝐴𝑀02 (1)
Figure 6. Comparison of CIECAM02 predictions with subjects estimations.
Figure 7. Comparison of iPhone (x-axis) with phones of LG-Nexus4 (top), Samsung (middle) and Huawei (bottom) by modified
It can be seen that after the correction using Eq. (1), the modified CIECAM02
can predict smartphones accurately.
Elham Khodamoradi Page 174
3.3 Comparison with other smart phones
After the modelling of iPhone using CIECAM02, a number of other smartphones
are evaluated as well, including, a LG-Nexus4, SamSung, and HuaWei. Figure 7
presents the comparison results by the calculations using CIECAM02 for all the
phones, whereas Figures 8 demonstrates a colour checker depicted on these
phones.
Figure 8. Colour checker depicted using four phones and CRT
When comparing with the other smartphones, in terms of hue, all three phones
tends to be more reddish for purplish colours than those displayed on an iPhone,
whereas the rest maintains near the same. With regard to lightness, for lighter
colours, all three phones unanimously appear darker than on iPhones with LG-
Nexus being the darkest with 25% darker, whereas 18% and 16% darker are
evidenced for HuaWei and SamSung respectively. However, the opposite
Elham Khodamoradi Page 175
phenomenon occurs when it comes to the representation of colourfulness. All
three phones of LG-Nexus4, SamSung, and HuaWei appear more colourful than
those displayed on an iPhone. For example, the colours on a LG-Nexus4 phone
appears systemically 10% more colourful than those depicted on an iPhone. In
addition, for both HuaWei and SamSung phones, they appear again to be 17%
and 22% more colourful, especially for colourful samples, the tendency
presenting across all three phones. Since these findings are based on only one
phone of each type, future study will focus on the investigation of large samples
with more similar phones.
Elham Khodamoradi Page 176
4. Conclusions
With 85% smartphone users perform their everyday tasks [Salesforce] using the
device due to its ability and connectivity, smartphone continues to revolutionize
how people perform their everyday activities. It is expected that this work will
contribute to this revolutionary and complement users with some information
when they perform online shopping buying colour incentive goods.
In summary, in terms of the measurement on each phone, encouragingly, the
variations among the same kind of phones are insignificant with less than 3 E
CIELAB units. In addition, when comparing with subjects’ hue estimations, all
phones and the CRT monitor appear to have similar hue values, indicating that
the hue values have been well reserved on those phones although some
variations among phones are observe red. However, when it is viewed on a
phone, a colour appears to show at variance with colourfulness by appearing
much more colourful. Furthermore, the correlation coefficient (r) for CIELUV L*
are 0.963, 0.959, 0.960, and 0.940 for iPhone, LG Nexus4, HuaWei and
SamSung respectively, and are 0.890, 0.876, 0.761, and 0.764 respectively for
CIELUV C* when comparing with the counterparts on the CRT monitor.
Therefore, the iPhone tends to be the best with fewer scatterings. To predict a
colour on a mobile phone using CIECAM02, the predictions rest on a number of
environmental parameters settings, e.g., f = 0.9, c = 0.59, and nc = 0.90, which
gives closer results with less scattering. Consequently, for an image with truthful
colour to be displayed on a mobile phone, the forward and reverse model of
CIECAM02 will have to be applied. Specifically, for iPhone5, nearly half of the
Elham Khodamoradi Page 177
colourfulness predicted by CIECAM02 need to be factored into. Further studies
are in place to take more number of phones into consideration.
Elham Khodamoradi Page 178
6. ACKNOWLEDGEMENTS
This research is in part funded by EU WIDTH project (2011-2014) under Marie
Curie Scheme. Their financial supported is gratefully acknowledged.
Elham Khodamoradi Page 179
11. REFERENCES
Bartleson, C.L., (1979), Changes in Color Appearance with Changes in Adaptation, Color Research & Application, Vol 4(3), pp119-138. DOI: 10.1002/col.5080040303
Bartleson, C.L., (1980), Measures of Brightness and Lightness, Die Farbe, Vol 28, No 3-6, pp132-148. ISSN: 0014-7680
Bartleson, C.L., and Breneman, E.l., (1967), Brightness Perception in Complex Fields, J. Opt. Soc. Am., Vol 57(7), pp953-975. doi.org/10.1364/JOSA.57.000953
Berlin, B. and Kay, P., (1969), Basic Color Terms: Their Universality and Evolution, Berkeley: University of California Press. ISBN: 1-57586-162-3
Berns, R.S., Billmeyer, F.W. and Saltzman, M., (2000), Principles of Colour Technology, 3rd Edition, John Wiley & Sons, New York. ISBN: 978-0-471-19459-0
Billmeyer, F.W Jr., Saltzman, M., (1981), Principles of Color Technology, 2nd Edition, John Wiley & Sons, New York. ISBN: 978-0471030522
Briggs, D., (1998) Modern Colour Theory for Digital and Traditional Painting. Julian Ashton Art School and National Art School, Sydney, Australia.
Broadbent, A.D., (2004), "A critical review of the development of the CIE1931 RGB color-matching functions". Color Research & Applications. Wiley Online Library. DOI: 10.1002/col.20020
Burnham, R.W., Hanes, R.M., Bartleson, C.J., (1963), Color: A Guide to Basic Facts and Concepts, John Wiley & Sons, New York. BCIN Number: 127430
CIE (1932), Commission lnternationale de I'Eclairage proceedings, Cambridge University Press.
CIE Publication 159. (2006), A colour appearance model for colour management systems, CIECAM02. DOI: 10.1002/col.20198
CIE Publication No. 17, (1970), International Lighting Vocabulary, Commission Internationale de l'Ec1airage, Paris. ISBN 978 3 900734 07 7
CIE: ISO 11664-1:2007(E) I CIE S 014-1/E:2006: Joint ISO/CIE Standard: CIE Colorimetry -Part 1: Standard Colorimetric Observers.
CIE: ISO 11664-4:2008(E)/CIE S 014-4/E:2007: Joint ISO/CIE Standard: CIE Colorimetry -Part 4: 1976 L *a*b* Colour Space.
CIE: ISO 11664-5:2009(E)/CIE S 014-5/E:2009: Joint ISO/CIE Standard: CIE Colorimetry-Part 5: CIE 1976 L*u*v* Colour Space and u', v' Uniform Chromaticity Scale Diagram.
Elham Khodamoradi Page 180
CIE: ISO 23603:2005(E)/CIE S 012/E:2004: Joint ISO/CIE Standard: Standard Method of Assessing the Spectral Quality of Daylight Simulators for Visual Appraisal and Measurement of Colour.
Davi, A., Fennessy, P., (1998), Digital Imaging for Photographers, 3rd Ed., Focal Press. ISBN-13: 978-0240515908
Drimbarean, A., and Whelan, P.F., (2001), Experiments in colour texture analysis, Pattern Recognition Letters, Vol 22(10), pp1161-1167. Elsevier Science Inc. New York. doi:10.1016/S0167-8655(01)00058-7
Evans, R.M., (1964), Variables of Perceived Color, J. Opt. Soc. Am. 54(12), pp1467-1474. PMID: 5868598
Fairchild and Lennie, (1992), Chromatic adaptation to natural and incandescent illuminants, Vol 32(11), pp2077-85. PMID:1304085.
Fairchild and Reniff, (1995), Time course of chromatic adaptation for colour-appearance judgments, Vol 12(5), pp824-33. PMID:7730950
Fairchild, M.D., (2005), Color Appearance Models, 2nd Ed., Wiley IS&T, Chichester, UK. ISBN 0-470-01216-1
Ford, A., Roberts, A., (1998), Colour Space Conversions, University of Westminster, London, UK. DOI: 10.3390/s130607838
Fraser B., Murphy, C., and Bunting, F., (2004) Real World Color Management: Strength Production Techniques, Peachit Press, Pearson Education, San Francisco. ISBN: 0-321-26722-2.
Gao, X., Wang, Y., Qian, Y., Gao, A., (2015), Modelling of Chromatic Contrast for Retrieval of Wallpaper Images, Color Research and Application, Vol 40(4), pp361-373. Wiley Periodicals. DOI: 10.1002/col.21897
Gevers, T., (2005), Combining color and shape information for illumination-viewpoint invariant object recognition. DOI: 10.1109/TIP.2005.860320. ISSN: 1057-7149
Gibson, N., (1967), The Measurement of Subjective Colour, Optica Acta, 14(3), pp219-243. Doi.org/10.1080/713818041
Gordon, J., Abramov, I., (1977), Color vision in the peripheral retina: II. Hue and saturation. Journal of the Optical Society of America, 67(2), pp202–207. PMID: 839300
Gorea, A. and Papathomas, T., (1991), Texture segregation by chromatic and achromatic visual pathways: an analogy with motion processing, Journal of the Optical Society of America A., Vol 8(2), pp386-393. doi.org/10.1364/JOSAA.8.000386
Grum, F., Becherer, R. J., (1979), Optical Radiation Measurements: Radiometry (Volume 1), F. Grum, Ed., Academic Press, New York. ISBN 13: 9780123049018
Elham Khodamoradi Page 181
Guild, J., (1932), The colorimetric properties of the spectrum, Philosophical Transactions of the Royal Society of London, Series A, Mathematical, Physical and Engineering Sciences. Vol 230, issue 681-693, pp 149–187. DOI: 10.1098/rsta.1932.0005
Hansen, T., Giesel, M., Gegenfurtner, K., (2008), Chromatic discrimination of nature objects, Journal of Vision, 8(1):2, pp1-19. doi:10.1167/8.1.2
Harris, A. C.; Weatherall, I. L. (1990). Objective evaluation of color variation in the sand-burrowing beetle Chaerodes trachyscelides White (Coleoptera: Tenebrionidae) by instrumental determination of CIE LAB values, Journal of the Royal Society of New Zealand. The Royal Society of New Zealand. 20 (3): 253–259. doi:10.1080/03036758.1990.10416819.
Helmhotz, H.V., (1924), Handbook of Physiological Optics, Vol.3, Ed., Southall, J.P.C., Optical Society of American, New York. ISBN 13: 9780486600161
Helson, H., Judd, D.B., Warren, M.H. (1952), Object-color Changes from daylight to incandescent Filament Illumination, Illumination Engineering, Vol 47, pp221-233.
Hering, E., (1878), Outline of a Theory of The Light Source, Translated by Hurvich, L. & Janeson, D., Harvard University Press, Massachusetts. ISBN-13: 978-0674649002.
Hunt, R.W., (1998). Measuring Colour (3rd Ed.), Ellis Horwood Ltd. ISBN 13: 9780745801254
Hunt, R.W.G. (1952), Light and Dark Adaptation and the Perception of Color, J. Opt. Soc. Am., Vol 42(3), pp190-199. doi.org/10.1364/JOSA.42.000190
Hunt, R.W.G., (1970), Measuring Colour, 2nd Ed., Ellis Horwood Ltd.
Hunt, R.W.G., (2004), The Reproduction of Colour, Wiley IS&T Series. ISBN: 978-0-470-02425-6
Hunt, R.W.G., (2011), Measuring colour, fourth edition. ISBN: 978-1-119-97537-3
Indow, T., Stevens, S.S., (1966), Scaling of Saturation and Hue, Perception & Psychophysjcs, Vol 1(4), pp253-271. doi: 10.3758/BF03207390
Ishak, I.G.H., Bouma, H., & van Bussel, J.J., (1970), Subjective Estimates of Colour Attributes for Surface Colours, Vision Research, Vol 10(6), pp489-500. doi.org/10.1364/JOSA.64.000743
Ishihara, S., (1998) Ishihara Tests for colour deficiency, Concise Edition, Kanehara Shuppan Co Ltd, Tokyo. ISBN-13: 978-0718600792
Judd, D. B. (1965), Progress report for OSA Committee on Uniform Color Scales. Die Farbe 14:287–295.
Julesz, B., (1962), Visual Pattern Discrimination, IRE Transactions on Information Theory, Vol 8(2), pp84-92. doi:10.1109/TIT.1962.1057698
Elham Khodamoradi Page 182
Julesz, B., (1981), Textons, the Elements of Texture Perception, and their Interactions, Nature, Vol 290, pp91-97. doi:10.1038/290091a0
Kelly, K.L., (1963), Lines of Constant Correlated Color Temperature Based on MacAdam's (u,v) Uniform Chromaticity Transformation of the CIE Diagram, Journal of the Optical Society of America, Vol 53(8). doi.org/10.1364/JOSA.53.000999
Luo M.R., Rigg, B., Hunt, R.W.G., Li, C., (2002), CMC 2000 Chromatic Adaptation Transform: CMCAT2000, Color Research and Application, Vol 27(1), pp49-58. DOI: 10.1002/col.10005
Luo, M.R, Gao, X.W., Scrivener, S.A.R., (1995), Quantifying Colour Appearance-Part V. Simultaneous Contrast, Color Research and Application, 20(1), pp18-28. DOI: 10.1002/col.5080200105
Luo, M.R., Clarke, A.A., Rhodes, R.A., Schappo, A., Scrivener, S.A.R., Tait, C.J., (1991b), Quantifying Colour Appearance, Part 11, Testing Colours Models Performance Using LUTCHI Colour Appearance Data, Color Research and Application, 16(3), pp181-197. DOI: 10.1002/col.5080160308
Luo, M.R., Clarke, A.A., Rhodes, R.A., Schappo, A., Scrivener, S.A.R., Tait, C.J., (1991a), Quantifying Colour Appearance, Part I, LUTCHI Colour Appearance Data, Color Research and Application, 16(3), pp166-179. DOI: 10.1002/col.5080160307
Luo, M.R., Gao, X.W., Rhodes, R.A., Xin, H.J., Clark, A.A., Scrivener, S.A.R., (1993c), Quantifying colour appearance - Part IV. Transmissive media, Color Research and Application, 18(3), pp191-206. DOI: 10.1002/col.5080180309
Luo, M.R., Gao, X.W., Rhodes, R.A., Xin, H.J., Clark, A.A., Scrivener, S.A.R., (1993b), Quantifying colour appearance - Part III. Supplementary LUTCHI colour appearance data, Color Research and Application, 18(2), pp. 98-113. DOI: 10.1002/col.5080180207
Luo, M.R., Gao, X.W., Scrivener, S. A. R., (1995), Quantifying Colour Appearance, Part V. Simultaneous Contrast, Color Research and Application, 20(1), pp18-28. DOI: 10.1002/col.5080200105
Moroney, N., Fairchild, M.D., Hunt, R.W.G., Li, C., Luo, M.R., Newman, T., (2002), The CIECAM02 Color Appearance Model, Society for Imaging Science and Technology - Tenth Color Imaging Conference Proceedings, pp. 23-27. ISBN 0-89208-241-0
Munsell Book of Colour. MPN: 40115B; UPC: 710762440074; UNSPSC: 60104813.
Nayatani, Y., Yamanaka, T., Sobagaki, H., (1972), Subjective Estimatation of Color Attributes for Surface Colours (Part 1: Reproducibility of Estimations), Acta Chromatica, 2, pp129-138.
Neaves, S.H. and Davidson, M.W., (1998), National High Magnetic Field Laboratory, The Florida State University.
Elham Khodamoradi Page 183
Newhall, S.M., Burnham, R.W., Clark, J.R., (1957), Comparison of Successive with Simultaneous Color Matching, J. Opt. Soc. Am., 47(1), pp43-56. doi.org/10.1364/JOSA.47.000043
Othman, A., Martinez, K., (2008), Colour appearance descriptors for image browsing and retrieval, Proc. SPIE Electronic Imaging: Multimedia Content Access: Algorithms and Systems, San Jose, USA. doi:10.1117/12.766882
Oulton, D.P., Peterman, E., and Andrew, W., (2008), Measuring the Contribution of Texture to Colour Appearance. International Conference 'Colour Science 98', Harrogate, Proc. Publ. by Leeds University.
Padgham, C.A., and Rowe, S.C.H., (1973), A Mass Colour Scaling Experiment, in Color 73, pp393-398, Adam Hilger, London.
Park, Y., Li C., Luo, M.R., (2006), Colour Appearance Modelling for Mobile Displays, Society for Imaging Science and Technology, 14th Color and Imaging Conference Proceedings, Scottsdale, Arizona pp18-23. ISSN 2169-2629
Pointer, M.R., (1981), A comparison of the CIE 1976 colour spaces, Color Research and Application, 6(2), pp108-118. DOI: 10.1002/col.5080060212
Pointer, M.R., (1980), The Concept of Colourfulness and Its Use for Deriving Grids for Assessing Colour Appearance, Color Research and Application, 5(2), pp99-107. DOI: 10.1002/col.5080050212
Qin-ling, D., (2012), Several Colour Appearance Phenomena in Colour Reproduction, 2nd International Conference on Electronic & Mechanical Engineering and Information Technology, Paris, France, Atlantis Press, pp1401-1404.
Qiu, G., (2002), Indexing chromatic and achromatic patterns for content-based colour image retrieval, Pattern Recognition, 35(8), pp1675-1686. doi.org/10.1016/S0031-3203(01)00162-5
Romero, Hernández-Andrés, Nieves & García, (2003), Color coordinates of objects with daylight changes, Color Research and Applics, 28(1), pp25-35. DOI: 10.1002/col.10111
Sangwine, S.J., Horne, R.E.N., (1998), The Colour Image Processing Handbook, Chapman & Hall. ISBN: 978-1-4613-7647-7
Schanda, J., (2007), The Future of Colorimetry in the CIE: Color Appearance, Colorimetry: Understanding the CIE System, John Wiley & Sons, Hoboken, New Jersey. ISBN 978-0-470-04904-4
Stevens, J.C., Stevens, S.S., (1963), Brightness Function: Effect of Adaptation, J. Opt. Soc Am., 53(3), pp375-385. doi.org/10.1364/JOSA.53.000375
Stevens, S.S. and Galanter, E.H., (1957), Ratio Scales and Category Scales for a Dozen Perceptual Continua, Exp. Psychol., 54(6), pp377-411. dx.doi.org/10.1037/h0043680
Elham Khodamoradi Page 184
Stitles, W.A., Wyszecki, G ., (2000), Color Science: Concepts and Methods, Quantitative Data and Formulae, 2nd ed., Wiley Classics Library. ISBN-10: 0471399183
Susstrunk, S., Buckley, R., Swen, S., (1999), Color and Imaging Conference, 7th Color and Imaging Conference Final Program and Proceedings, pp. 127-134(8). ISSN 2166-9635
Webste, M.A., Mallon, R.J.D., (1997), Adaptation and colour statistics of natural images, Vision Research 37(23), pp3283-3298. doi.org/10.1016/S0042-6989(97)00125-9
Wright, W.D., (1928), A re-determination of the trichromatic coefficients of the spectral colours, Transactions of the Optical Society, 30(4), pp141-164. iopscience.iop.org/1475-4878/30/4/301
Wu, R.C., Wardman, R.H., (2007), Proposed modification to the CIECAM02 colour appearance model to include the simultaneous contrast effects, Color Res. Appl., 32(2), pp121-129. DOI: 10.1002/col.20297
Wu, R.C., Wardman, R.H., (2007), Proposed modification to the CIECAM02 colour appearance model to include the simultaneous contrast effects, Color Research and Application, 32(2), pp121-129. DOI: 10.1002/col.20297
Wyszecki, G. & Stiles, W.S., (1982), Color Science: Concepts and Methods, Quantitative Data and Formulae, Second Edition, Wiley lnterscience, N.Y. ISBN: 978-0-471-39918-6
Wyszecki, G., (1973), Current Developments in Colorimetry, Colour 73, pp21-50, Adam Hilger, London.
Young, T., (1802), The Bakerian Lecture: On the Theory of Light and Colours, Philosophical Transactions of the Royal Society of London, 92(1802), pp12-48. www.jstor.org/stable/107113
Elham Khodamoradi Page 185
WEB REFRENCES
App market: http://www.pharmaphorum.com/articles/moving-into-the-mobile-app-market.
Apple: http://www.gsmarena.com/apple
Apple:http://developer.apple.com/documentation/QuickTime/REF/refVectors.21.htm#pgfI=20708) , 1999.
CIE 1931 colour space: http://en.wikipedia.org/wiki/CIE_1931_color_space. /. Retrieved 20 Oct. 2015.
CIE: http://www.cie.co.at/
Cisco: http://www.google.co .uklcisco, /. Retrieved Apri1 2015. Pic 32
CISCO:http://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/VNI_Hyperconnectivity_WP.html
Color, Wikipedia: http://en.wikipedia.org/wiki/Color, /. Retrieved September2010.
Color: http://en.wikipedia.org/wiki/Color, /. Retrieved 10 Septermber2010.
Colour definition: http://www.wordreference.com/definition/colour. Retrieved September 2010.
Counter Point Research:http://www.counterpointresearch.com/top10./. Retrieved in February 2015.
Digital stats: http://digital-stats. blogspot. co. uk/20 12/0 1/ownership-usage-of-mobile-consumer. html/. Retrieved March 2015-pic 33
Digital-stats: http://digital-stats.blogspot.co.uk/2012/01/ownership-usage-of-mobile-consumer.html
Emarket: http://www.counterpointresearch.com/top1 0/. Retrieved February 2015.
facweb; http://facweb.cs.depaul.edu/sgrais/color_context. Retrieved September, 2016
Flickr: https://www.flickr.com/cameras/. Retrieved February, 2015.
healthcare-apps-medical-devices I/. Retrieved March 2015-pic 20.
http://dba.med.sc.edu/price/irf/Adobe_tg/color/variables.html
http://dba.med.sc.edu/price/irf/Adobe_tg/models/ciexyz.html
index-vniNNI_Hyperconnectivity_WP.html/. Retrieved March 2015-pic 17.
IPCarrier:http://ipcarrier.blogspot.co.uk/2012/11/fighting-dumb-pipe-wont-work.html#!/2012/11/fighting-dumb-pipe-wont-work.html
Elham Khodamoradi Page 186
iPhone camera:https://www.google.eo.uk/search?q=iphone+5+camera/. Retrieved April 2015.
Lipcarrier: http:/lipcarrier.blogspot.co.uk/. Retrieved March 2015-pic 31.
lntor refreluminous-landscape [Online]: http://www.luminous-I
marketing-statistics /. Retrieved March 2015 pic 15.
Mobile master: http:l/www. mobilemastinfo.com/stats-and-facts/. Retrieved April 2015
Mobile usage chart: https://www.google.eo.uk/search? mobile+usage+uk+chart I /. Retrieved April 2015.
one_5-news-4796.php/. Retrieved March 2015-pic 23.
Pharmaphorum:http://www.pharmaphorum.com/articles/moving-into-the-mobile-app-market-are-healthcare-apps-medical-devices /. Retrieved March 2015- pic 34
Researchgat :www.researchgate.net/publication/227991182_CIE_Color_Appearance_Models_and_Associated_Color_Spaces /. Retrieved March 2015
SmartInSight:http://www.smartinsights.com/mobile-marketing/mobile-marketing-analytics/mobile-marketing-statistics/. Retrieved March 2015-pic 30
Smartphone trend:http://trends. e-strategyblog. com/2014/04/28/smart-phone-user -penetration-in-westerneurope-by-country-2012-2017/18661/. Retrieved March 2015-pic 18
Smartphone: https://www.google.eo.uk/ smartphonel/. Retrieved April 2015
Smartphone:http://www. portio research. com/blog/20 13/0 1/the-growing-uk-smartphone-market.
Smartphone:http://www.smartinsights.com/mobile-marketing/mobile-marketing-analytics/mobile/. Retrieved April 2015.
Spyder: http://spyder.datacolor.com/. Retrieved September 2010.
VisTex: http://vismod.media.mit.edu/vismod/imagery/VisionTexture/vistex.html
Walk digital: http:llwww.walkdigital.com/mobile-stats/. Retrieved April 2015.
We are social: http://wearesocial.net/. Retrieved February 2015.
www.studenthealth.ucla.edu/FormsDocuments/LayereoftheHumanEye)
www.wolfcrow.com/blog/what-is-color-space-and-the-tristimulus-values-xyz